@omote/core 0.1.2 → 0.2.0

package/dist/index.js

@@ -5982,7 +5982,7 @@ async function load_video(src, { num_frames = null, fps = null } = {}) {
   video.remove();
   return new RawVideo(frames, duration);
 }
-var ONNX_WEB, import_meta, __defProp2, __export2, emptyObj, node_fs_default, emptyObj2, node_path_default, emptyObj3, node_url_default, VERSION, IS_PROCESS_AVAILABLE, IS_NODE_ENV, IS_FS_AVAILABLE, IS_PATH_AVAILABLE, IS_DENO_RUNTIME, IS_BUN_RUNTIME, IS_BROWSER_ENV, IS_WEBWORKER_ENV, IS_WEB_CACHE_AVAILABLE, IS_WEBGPU_AVAILABLE, IS_WEBNN_AVAILABLE, isSafari, IS_SAFARI, apis, RUNNING_LOCALLY, dirname__, DEFAULT_CACHE_DIR, DEFAULT_LOCAL_MODEL_PATH, localModelPath, env2, Callable, CONTENT_TYPE_MAP, FileResponse, FileCache, ERROR_MAPPING, MAX_EXTERNAL_DATA_CHUNKS, REPO_ID_REGEX, P2FFT, NP2FFT, FFT, uint16_to_float32, onnxruntime_node_exports, noop, emptyObj4, onnxruntime_node_default, Readable, pipeline, createWriteStream, createReadStream, DEVICE_TO_EXECUTION_PROVIDER_MAPPING, LOG_LEVELS, DEFAULT_LOG_LEVEL, supportedDevices, defaultDevices, ONNX, ORT_SYMBOL, InferenceSession2, IS_WEB_ENV, webInitChain, wasmLoadPromise, webInferenceChain, ONNX_ENV, wrap, _a, TensorOpRegistry, DataTypeMap, Tensor22, arrayToIndexTensor, PriorityQueue, CharTrie, CharTrieNode, TokenLattice, TokenLatticeNode, DictionarySplitter, LRUCache, TOKEN_TYPES, Token, ORDERED_MAPPING_TABLE, ESCAPE_CHARACTERS, Statement, Program, If, For, Break, Continue, SetStatement, Macro, Comment, Expression, MemberExpression, CallExpression, Identifier, Literal, IntegerLiteral, FloatLiteral, StringLiteral, ArrayLiteral, TupleLiteral, ObjectLiteral, BinaryExpression, FilterExpression, FilterStatement, SelectExpression, TestExpression, UnaryExpression, SliceExpression, KeywordArgumentExpression, SpreadExpression, CallStatement, Ternary, BreakControl, ContinueControl, RuntimeValue, IntegerValue, FloatValue, StringValue, BooleanValue, ObjectValue, KeywordArgumentsValue, ArrayValue, TupleValue, FunctionValue, NullValue, UndefinedValue, Environment, Interpreter, NEWLINE, OPEN_STATEMENT, CLOSE_STATEMENT, Template, WHISPER_LANGUAGES, WHISPER_LANGUAGE_MAPPING, WHISPER_TO_LANGUAGE_CODE_MAPPING, PUNCTUATION_REGEX, PUNCTUATION_ONLY_REGEX, BLOOM_SPLIT_CHARS, PROBLEMATIC_REGEX_MAP, AddedToken, TokenizerModel, WordPieceTokenizer, Unigram, BYTES_TO_UNICODE, UNICODE_TO_BYTES, BPE, LegacyTokenizerModel, Normalizer, Replace, UnicodeNormalizer, NFC, NFD, NFKC, NFKD, StripNormalizer, StripAccents, Lowercase, Prepend, NormalizerSequence, BertNormalizer, PreTokenizer, BertPreTokenizer, ByteLevelPreTokenizer, SplitPreTokenizer, PunctuationPreTokenizer, DigitsPreTokenizer, PostProcessor, BertProcessing, RobertaProcessing, TemplateProcessing, ByteLevelPostProcessor, PostProcessorSequence, Decoder, ReplaceDecoder, ByteFallback, FuseDecoder, StripDecoder, WordPieceDecoder, ByteLevelDecoder, CTCDecoder, DecoderSequence, BPEDecoder, VitsDecoder, MetaspacePreTokenizer, MetaspaceDecoder, Precompiled, PreTokenizerSequence, WhitespacePreTokenizer, WhitespaceSplit, ReplacePreTokenizer, FixedLengthPreTokenizer, SPECIAL_TOKEN_ATTRIBUTES, PreTrainedTokenizer, TokenizersBackend, BertTokenizer, AlbertTokenizer, MobileBertTokenizer, SqueezeBertTokenizer, DebertaTokenizer, DebertaV2Tokenizer, HerbertTokenizer, ConvBertTokenizer, RoFormerTokenizer, DistilBertTokenizer, CamembertTokenizer, XLMTokenizer, ElectraTokenizer, T5Tokenizer, GPT2Tokenizer, BartTokenizer, MBartTokenizer, MBart50Tokenizer, RobertaTokenizer, BloomTokenizer, SPIECE_UNDERLINE, LlamaTokenizer, CodeLlamaTokenizer, XLMRobertaTokenizer, MPNetTokenizer, FalconTokenizer, GPTNeoXTokenizer, EsmTokenizer, Qwen2Tokenizer, GemmaTokenizer, Grok1Tokenizer, NllbTokenizer, M2M100Tokenizer, WhisperTokenizer, CodeGenTokenizer, CLIPTokenizer, SiglipTokenizer, MarianTokenizer, Wav2Vec2CTCTokenizer, BlenderbotTokenizer, BlenderbotSmallTokenizer, SpeechT5Tokenizer, NougatTokenizer, VitsTokenizer, CohereTokenizer, MgpstrTokenizer, Ernie4_5_Tokenizer, _a2, AutoTokenizer, GITHUB_ISSUE_URL, FEATURE_EXTRACTOR_NAME, IMAGE_PROCESSOR_NAME, PROCESSOR_NAME, CHAT_TEMPLATE_NAME, _a3, Processor, processors_exports, FeatureExtractor, feature_extractors_exports, noop2, Readable2, noop3, pipeline2, HERTZ_TO_MEL_MAPPING, MEL_TO_HERTZ_MAPPING, RawAudio, ASTFeatureExtractor, EncodecFeatureExtractor, ChatterboxFeatureExtractor, ClapFeatureExtractor, DacFeatureExtractor, Gemma3nAudioFeatureExtractor, MoonshineFeatureExtractor, EPSILON, ParakeetFeatureExtractor, PyAnnoteFeatureExtractor, SeamlessM4TFeatureExtractor, SnacFeatureExtractor, SpeechT5FeatureExtractor, Wav2Vec2FeatureExtractor, WeSpeakerFeatureExtractor, WhisperFeatureExtractor, emptyObj5, sharp_default, createCanvasFunction, ImageDataClass, loadImageFunction, IS_BROWSER_OR_WEBWORKER, RESAMPLING_MAPPING, CONTENT_TYPE_MAP2, RawImage, load_image, ImageProcessor, AutoFeatureExtractor, _a4, ChatterboxProcessor, image_processors_exports, BeitFeatureExtractor, BitImageProcessor, ChineseCLIPFeatureExtractor, CLIPImageProcessor, CLIPFeatureExtractor, ConvNextImageProcessor, ConvNextFeatureExtractor, DeiTImageProcessor, DeiTFeatureExtractor, DetrImageProcessor, DetrFeatureExtractor, DINOv3ViTImageProcessor, DonutImageProcessor, DonutFeatureExtractor, DPTImageProcessor, DPTFeatureExtractor, EfficientNetImageProcessor, GLPNFeatureExtractor, GroundingDinoImageProcessor, Idefics3ImageProcessor, VLMImageProcessor, JinaCLIPImageProcessor, LlavaOnevisionImageProcessor, MaskFormerImageProcessor, MaskFormerFeatureExtractor, Mask2FormerImageProcessor, MobileNetV1ImageProcessor, MobileNetV1FeatureExtractor, MobileNetV2ImageProcessor, MobileNetV2FeatureExtractor, MobileNetV3ImageProcessor, MobileNetV3FeatureExtractor, MobileNetV4ImageProcessor, MobileNetV4FeatureExtractor, MobileViTImageProcessor, MobileViTFeatureExtractor, NougatImageProcessor, OwlViTImageProcessor, OwlViTFeatureExtractor, Owlv2ImageProcessor, IMAGE_SIZE, SLICE_AXES, ceil, floor, sqrt, Phi3VImageProcessor, PvtImageProcessor, Qwen2VLImageProcessor, RTDetrImageProcessor, SamImageProcessor, SegformerImageProcessor, SegformerFeatureExtractor, SiglipImageProcessor, Swin2SRImageProcessor, ViTImageProcessor, ViTFeatureExtractor, VitMatteImageProcessor, VitPoseImageProcessor, YolosImageProcessor, YolosFeatureExtractor, AutoImageProcessor, _a5, Florence2Processor, _a6, Gemma3nProcessor, _a7, GroundingDinoProcessor, _a8, Idefics3Processor, _a9, VLChatProcessor, _a10, JinaCLIPProcessor, _a11, LlavaProcessor, DECODE_TYPE_MAPPING, _a12, MgpstrProcessor, _a13, MoonshineProcessor, _a14, OwlViTProcessor, IMAGE_TOKEN, IMAGE_TOKEN_PATTERN, _a15, Phi3VProcessor, IMAGE_TOKEN2, _a16, PaliGemmaProcessor, _a17, PyAnnoteProcessor, _a18, Qwen2VLProcessor, _a19, SamProcessor, Sam2Processor, Sam2VideoProcessor, _a20, SpeechT5Processor, _a21, UltravoxProcessor, AUDIO_TOKEN, BEGIN_AUDIO_TOKEN, NUM_AUDIO_TOKENS, _a22, VoxtralProcessor, _a23, Wav2Vec2Processor, _a24, Wav2Vec2ProcessorWithLM, _a25, WhisperProcessor, AutoProcessor, PretrainedConfig, AutoConfig, DEVICE_TYPES, isWebGpuFp16Supported, DATA_TYPES, DEFAULT_DEVICE_DTYPE_MAPPING, DEFAULT_DTYPE_SUFFIX_MAPPING, LogitsProcessor, LogitsWarper, LogitsProcessorList, ForcedBOSTokenLogitsProcessor, ForcedEOSTokenLogitsProcessor, SuppressTokensAtBeginLogitsProcessor, WhisperTimeStampLogitsProcessor, NoRepeatNGramLogitsProcessor, RepetitionPenaltyLogitsProcessor, MinLengthLogitsProcessor, MinNewTokensLengthLogitsProcessor, NoBadWordsLogitsProcessor, ClassifierFreeGuidanceLogitsProcessor, TemperatureLogitsWarper, TopPLogitsWarper, TopKLogitsWarper, GenerationConfig, StoppingCriteria, StoppingCriteriaList, MaxLengthCriteria, EosTokenCriteria, InterruptableStoppingCriteria, LogitsSampler, GreedySampler, MultinomialSampler, BeamSearchSampler, WhisperGenerationConfig, MODEL_TYPES, MODEL_TYPE_MAPPING, MODEL_NAME_TO_CLASS_MAPPING, MODEL_CLASS_TO_NAME_MAPPING, PreTrainedModel, ModelOutput, BaseModelOutput, BertPreTrainedModel, BertModel, BertForMaskedLM, BertForSequenceClassification, BertForTokenClassification, BertForQuestionAnswering, NeoBertPreTrainedModel, NeoBertModel, NeoBertForMaskedLM, NeoBertForSequenceClassification, NeoBertForTokenClassification, NeoBertForQuestionAnswering, ModernBertPreTrainedModel, ModernBertModel, ModernBertForMaskedLM, ModernBertForSequenceClassification, ModernBertForTokenClassification, ModernBertDecoderPreTrainedModel, ModernBertDecoderModel, ModernBertDecoderForCausalLM, NomicBertPreTrainedModel, NomicBertModel, RoFormerPreTrainedModel, RoFormerModel, RoFormerForMaskedLM, RoFormerForSequenceClassification, RoFormerForTokenClassification, RoFormerForQuestionAnswering, ConvBertPreTrainedModel, ConvBertModel, ConvBertForMaskedLM, ConvBertForSequenceClassification, ConvBertForTokenClassification, ConvBertForQuestionAnswering, ElectraPreTrainedModel, ElectraModel, ElectraForMaskedLM, ElectraForSequenceClassification, ElectraForTokenClassification, ElectraForQuestionAnswering, CamembertPreTrainedModel, CamembertModel, CamembertForMaskedLM, CamembertForSequenceClassification, CamembertForTokenClassification, CamembertForQuestionAnswering, DebertaPreTrainedModel, DebertaModel, DebertaForMaskedLM, DebertaForSequenceClassification, DebertaForTokenClassification, DebertaForQuestionAnswering, DebertaV2PreTrainedModel, DebertaV2Model, DebertaV2ForMaskedLM, DebertaV2ForSequenceClassification, DebertaV2ForTokenClassification, DebertaV2ForQuestionAnswering, DistilBertPreTrainedModel, DistilBertModel, DistilBertForSequenceClassification, DistilBertForTokenClassification, DistilBertForQuestionAnswering, DistilBertForMaskedLM, EsmPreTrainedModel, EsmModel, EsmForMaskedLM, EsmForSequenceClassification, EsmForTokenClassification, MobileBertPreTrainedModel, MobileBertModel, MobileBertForMaskedLM, MobileBertForSequenceClassification, MobileBertForQuestionAnswering, MPNetPreTrainedModel, MPNetModel, MPNetForMaskedLM, MPNetForSequenceClassification, MPNetForTokenClassification, MPNetForQuestionAnswering, SqueezeBertPreTrainedModel, SqueezeBertModel, SqueezeBertForMaskedLM, SqueezeBertForSequenceClassification, SqueezeBertForQuestionAnswering, AlbertPreTrainedModel, AlbertModel, AlbertForSequenceClassification, AlbertForQuestionAnswering, AlbertForMaskedLM, T5PreTrainedModel, T5Model, T5ForConditionalGeneration, LongT5PreTrainedModel, LongT5Model, LongT5ForConditionalGeneration, MT5PreTrainedModel, MT5Model, MT5ForConditionalGeneration, BartPretrainedModel, BartModel, BartForConditionalGeneration, BartForSequenceClassification, MBartPreTrainedModel, MBartModel, MBartForConditionalGeneration, MBartForSequenceClassification, MBartForCausalLM, BlenderbotPreTrainedModel, BlenderbotModel, BlenderbotForConditionalGeneration, BlenderbotSmallPreTrainedModel, BlenderbotSmallModel, BlenderbotSmallForConditionalGeneration, RobertaPreTrainedModel, RobertaModel, RobertaForMaskedLM, RobertaForSequenceClassification, RobertaForTokenClassification, RobertaForQuestionAnswering, XLMPreTrainedModel, XLMModel, XLMWithLMHeadModel, XLMForSequenceClassification, XLMForTokenClassification, XLMForQuestionAnswering, XLMRobertaPreTrainedModel, XLMRobertaModel, XLMRobertaForMaskedLM, XLMRobertaForSequenceClassification, XLMRobertaForTokenClassification, XLMRobertaForQuestionAnswering, ASTPreTrainedModel, ASTModel, ASTForAudioClassification, WhisperPreTrainedModel, WhisperModel, WhisperForConditionalGeneration, LiteWhisperForConditionalGeneration, MoonshinePreTrainedModel, MoonshineModel, MoonshineForConditionalGeneration, VisionEncoderDecoderModel, LlavaPreTrainedModel, LlavaForConditionalGeneration, LlavaOnevisionForConditionalGeneration, Moondream1ForConditionalGeneration, Florence2PreTrainedModel, Florence2ForConditionalGeneration, PaliGemmaPreTrainedModel, PaliGemmaForConditionalGeneration, LlavaQwen2ForCausalLM, Gemma3nPreTrainedModel, Gemma3nForConditionalGeneration, Idefics3PreTrainedModel, Idefics3ForConditionalGeneration, SmolVLMForConditionalGeneration, Phi3VPreTrainedModel, Phi3VForCausalLM, CLIPPreTrainedModel, CLIPModel, CLIPTextModel, CLIPTextModelWithProjection, CLIPVisionModel, CLIPVisionModelWithProjection, SiglipPreTrainedModel, SiglipModel, SiglipTextModel, SiglipVisionModel, ChineseCLIPPreTrainedModel, ChineseCLIPModel, JinaCLIPPreTrainedModel, JinaCLIPModel, JinaCLIPTextModel, JinaCLIPVisionModel, CLIPSegPreTrainedModel, CLIPSegModel, CLIPSegForImageSegmentation, GPT2PreTrainedModel, GPT2Model, GPT2LMHeadModel, GptOssPreTrainedModel, GptOssModel, GptOssForCausalLM, JAISPreTrainedModel, JAISModel, JAISLMHeadModel, GPTNeoPreTrainedModel, GPTNeoModel, GPTNeoForCausalLM, GPTNeoXPreTrainedModel, GPTNeoXModel, GPTNeoXForCausalLM, GPTJPreTrainedModel, GPTJModel, GPTJForCausalLM, GPTBigCodePreTrainedModel, GPTBigCodeModel, GPTBigCodeForCausalLM, CodeGenPreTrainedModel, CodeGenModel, CodeGenForCausalLM, LlamaPreTrainedModel, LlamaModel, LlamaForCausalLM, Llama4PreTrainedModel, Llama4ForCausalLM, NanoChatPreTrainedModel, NanoChatModel, NanoChatForCausalLM, ApertusPreTrainedModel, ApertusModel, ApertusForCausalLM, ArceePreTrainedModel, ArceeModel, ArceeForCausalLM, Lfm2PreTrainedModel, Lfm2Model, Lfm2ForCausalLM, SmolLM3PreTrainedModel, SmolLM3Model, SmolLM3ForCausalLM, HeliumPreTrainedModel, HeliumModel, HeliumForCausalLM, GlmPreTrainedModel, GlmModel, GlmForCausalLM, ExaonePreTrainedModel, ExaoneModel, ExaoneForCausalLM, MobileLLMPreTrainedModel, MobileLLMModel, MobileLLMForCausalLM, OlmoPreTrainedModel, OlmoModel, OlmoForCausalLM, Olmo2PreTrainedModel, Olmo2Model, Olmo2ForCausalLM, Olmo3PreTrainedModel, Olmo3Model, Olmo3ForCausalLM, GranitePreTrainedModel, GraniteModel, GraniteForCausalLM, GraniteMoeHybridPreTrainedModel, GraniteMoeHybridModel, GraniteMoeHybridForCausalLM, CoherePreTrainedModel, CohereModel, CohereForCausalLM, GemmaPreTrainedModel, GemmaModel, GemmaForCausalLM, Gemma2PreTrainedModel, Gemma2Model, Gemma2ForCausalLM, VaultGemmaPreTrainedModel, VaultGemmaModel, VaultGemmaForCausalLM, Gemma3PreTrainedModel, Gemma3Model, Gemma3ForCausalLM, OpenELMPreTrainedModel, OpenELMModel, OpenELMForCausalLM, Qwen2PreTrainedModel, Qwen2Model, Qwen2ForCausalLM, Qwen3PreTrainedModel, Qwen3Model, Qwen3ForCausalLM, Qwen2VLPreTrainedModel, Qwen2VLForConditionalGeneration, PhiPreTrainedModel, PhiModel, PhiForCausalLM, Phi3PreTrainedModel, Phi3Model, Phi3ForCausalLM, BloomPreTrainedModel, BloomModel, BloomForCausalLM, MptPreTrainedModel, MptModel, MptForCausalLM, OPTPreTrainedModel, OPTModel, OPTForCausalLM, ViTPreTrainedModel, ViTModel, ViTForImageClassification, IJepaPreTrainedModel, IJepaModel, IJepaForImageClassification, VitPosePreTrainedModel, VitPoseForPoseEstimation, PvtPreTrainedModel, PvtModel, PvtForImageClassification, ViTMAEPreTrainedModel, ViTMAEModel, ViTMSNPreTrainedModel, ViTMSNModel, ViTMSNForImageClassification, GroupViTPreTrainedModel, GroupViTModel, FastViTPreTrainedModel, FastViTModel, FastViTForImageClassification, VitMattePreTrainedModel, VitMatteForImageMatting, MobileViTPreTrainedModel, MobileViTModel, MobileViTForImageClassification, MobileViTV2PreTrainedModel, MobileViTV2Model, MobileViTV2ForImageClassification, OwlViTPreTrainedModel, OwlViTModel, OwlViTForObjectDetection, Owlv2PreTrainedModel, Owlv2Model, Owlv2ForObjectDetection, BeitPreTrainedModel, BeitModel, BeitForImageClassification, DetrPreTrainedModel, DetrModel, DetrForObjectDetection, DetrForSegmentation, DetrObjectDetectionOutput, DetrSegmentationOutput, RTDetrPreTrainedModel, RTDetrModel, RTDetrForObjectDetection, RTDetrObjectDetectionOutput, RTDetrV2PreTrainedModel, RTDetrV2Model, RTDetrV2ForObjectDetection, RTDetrV2ObjectDetectionOutput, RFDetrPreTrainedModel, RFDetrModel, RFDetrForObjectDetection, RFDetrObjectDetectionOutput, DFinePreTrainedModel, DFineModel, DFineForObjectDetection, TableTransformerPreTrainedModel, TableTransformerModel, TableTransformerForObjectDetection, TableTransformerObjectDetectionOutput, DeiTPreTrainedModel, DeiTModel, DeiTForImageClassification, HieraPreTrainedModel, HieraModel, HieraForImageClassification, ResNetPreTrainedModel, ResNetModel, ResNetForImageClassification, SwinPreTrainedModel, SwinModel, SwinForImageClassification, SwinForSemanticSegmentation, Swin2SRPreTrainedModel, Swin2SRModel, Swin2SRForImageSuperResolution, DPTPreTrainedModel, DPTModel, DPTForDepthEstimation, DepthAnythingPreTrainedModel, DepthAnythingForDepthEstimation, SapiensPreTrainedModel, SapiensForSemanticSegmentation, SapiensForDepthEstimation, SapiensForNormalEstimation, DepthProPreTrainedModel, DepthProForDepthEstimation, Metric3DPreTrainedModel, Metric3DForDepthEstimation, Metric3Dv2PreTrainedModel, Metric3Dv2ForDepthEstimation, MaskFormerPreTrainedModel, MaskFormerModel, MaskFormerForInstanceSegmentation, GLPNPreTrainedModel, GLPNModel, GLPNForDepthEstimation, DonutSwinPreTrainedModel, DonutSwinModel, ConvNextPreTrainedModel, ConvNextModel, ConvNextForImageClassification, ConvNextV2PreTrainedModel, ConvNextV2Model, ConvNextV2ForImageClassification, Dinov2PreTrainedModel, Dinov2Model, Dinov2ForImageClassification, Dinov2WithRegistersPreTrainedModel, Dinov2WithRegistersModel, Dinov2WithRegistersForImageClassification, DINOv3ViTPreTrainedModel, DINOv3ViTModel, DINOv3ConvNextPreTrainedModel, DINOv3ConvNextModel, GroundingDinoPreTrainedModel, GroundingDinoForObjectDetection, YolosPreTrainedModel, YolosModel, YolosForObjectDetection, YolosObjectDetectionOutput, SamPreTrainedModel, SamModel, SamImageSegmentationOutput, Sam2ImageSegmentationOutput, Sam2PreTrainedModel, Sam2Model, EdgeTamModel, Sam3TrackerModel, MarianPreTrainedModel, MarianModel, MarianMTModel, M2M100PreTrainedModel, M2M100Model, M2M100ForConditionalGeneration, Wav2Vec2PreTrainedModel, Wav2Vec2Model, Wav2Vec2ForCTC, Wav2Vec2ForSequenceClassification, Wav2Vec2ForAudioFrameClassification, ParakeetPreTrainedModel, ParakeetForCTC, PyAnnotePreTrainedModel, PyAnnoteModel, PyAnnoteForAudioFrameClassification, WeSpeakerResNetPreTrainedModel, WeSpeakerResNetModel, UniSpeechPreTrainedModel, UniSpeechModel, UniSpeechForCTC, UniSpeechForSequenceClassification, UniSpeechSatPreTrainedModel, UniSpeechSatModel, UniSpeechSatForCTC, UniSpeechSatForSequenceClassification, UniSpeechSatForAudioFrameClassification, Wav2Vec2BertPreTrainedModel, Wav2Vec2BertModel, Wav2Vec2BertForCTC, Wav2Vec2BertForSequenceClassification, HubertPreTrainedModel, HubertModel, HubertForCTC, HubertForSequenceClassification, WavLMPreTrainedModel, WavLMModel, WavLMForCTC, WavLMForSequenceClassification, WavLMForXVector, WavLMForAudioFrameClassification, StyleTextToSpeech2PreTrainedModel, StyleTextToSpeech2Model, SpeechT5PreTrainedModel, SpeechT5Model, SpeechT5ForSpeechToText, SpeechT5ForTextToSpeech, SpeechT5HifiGan, SupertonicPreTrainedModel, SupertonicForConditionalGeneration, TrOCRPreTrainedModel, TrOCRForCausalLM, MistralPreTrainedModel, MistralModel, MistralForCausalLM, Ernie4_5_PretrainedModel, Ernie4_5_Model, Ernie4_5_ForCausalLM, Starcoder2PreTrainedModel, Starcoder2Model, Starcoder2ForCausalLM, FalconPreTrainedModel, FalconModel, FalconForCausalLM, ClapPreTrainedModel, ClapModel, ClapTextModelWithProjection, ClapAudioModelWithProjection, VitsPreTrainedModel, VitsModel, SegformerPreTrainedModel, SegformerModel, SegformerForImageClassification, SegformerForSemanticSegmentation, StableLmPreTrainedModel, StableLmModel, StableLmForCausalLM, EfficientNetPreTrainedModel, EfficientNetModel, EfficientNetForImageClassification, MusicgenPreTrainedModel, MusicgenModel, MusicgenForCausalLM, MusicgenForConditionalGeneration, MobileNetV1PreTrainedModel, MobileNetV1Model, MobileNetV1ForImageClassification, MobileNetV1ForSemanticSegmentation, MobileNetV2PreTrainedModel, MobileNetV2Model, MobileNetV2ForImageClassification, MobileNetV2ForSemanticSegmentation, MobileNetV3PreTrainedModel, MobileNetV3Model, MobileNetV3ForImageClassification, MobileNetV3ForSemanticSegmentation, MobileNetV4PreTrainedModel, MobileNetV4Model, MobileNetV4ForImageClassification, MobileNetV4ForSemanticSegmentation, DecisionTransformerPreTrainedModel, DecisionTransformerModel, MultiModalityPreTrainedModel, MultiModalityCausalLM, MgpstrModelOutput, MgpstrPreTrainedModel, MgpstrForSceneTextRecognition, PatchTSTPreTrainedModel, PatchTSTModel, PatchTSTForPrediction, PatchTSMixerPreTrainedModel, PatchTSMixerModel, PatchTSMixerForPrediction, UltravoxPreTrainedModel, UltravoxModel, VoxtralForConditionalGeneration, MimiPreTrainedModel, MimiEncoderOutput, MimiDecoderOutput, MimiModel, MimiEncoderModel, MimiDecoderModel, DacPreTrainedModel, DacEncoderOutput, DacDecoderOutput, DacModel, DacEncoderModel, DacDecoderModel, SnacPreTrainedModel, SnacModel, SnacEncoderModel, SnacDecoderModel, ChatterboxPreTrainedModel, ChatterboxModel, _a26, PretrainedMixin, MODEL_MAPPING_NAMES_ENCODER_ONLY, MODEL_MAPPING_NAMES_ENCODER_DECODER, MODEL_MAPPING_NAMES_AUTO_ENCODER, MODEL_MAPPING_NAMES_DECODER_ONLY, MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES, MODEL_FOR_TEXT_TO_SPECTROGRAM_MAPPING_NAMES, MODEL_FOR_TEXT_TO_WAVEFORM_MAPPING_NAMES, MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES, MODEL_FOR_CAUSAL_LM_MAPPING_NAMES, MODEL_FOR_MULTIMODALITY_MAPPING_NAMES, MODEL_FOR_MASKED_LM_MAPPING_NAMES, MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES, MODEL_FOR_VISION_2_SEQ_MAPPING_NAMES, MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES, MODEL_FOR_AUDIO_TEXT_TO_TEXT_MAPPING_NAMES, MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES, MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES, MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING_NAMES, MODEL_FOR_IMAGE_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_UNIVERSAL_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_MASK_GENERATION_MAPPING_NAMES, MODEL_FOR_CTC_MAPPING_NAMES, MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_AUDIO_XVECTOR_MAPPING_NAMES, MODEL_FOR_AUDIO_FRAME_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_IMAGE_MATTING_MAPPING_NAMES, MODEL_FOR_TIME_SERIES_PREDICTION_MAPPING_NAMES, MODEL_FOR_IMAGE_TO_IMAGE_MAPPING_NAMES, MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES, MODEL_FOR_NORMAL_ESTIMATION_MAPPING_NAMES, MODEL_FOR_POSE_ESTIMATION_MAPPING_NAMES, MODEL_FOR_IMAGE_FEATURE_EXTRACTION_MAPPING_NAMES, MODEL_CLASS_TYPE_MAPPING, CUSTOM_MAPPING, CUSTOM_ARCHITECTURES, _a27, AutoModel, _a28, AutoModelForSequenceClassification, _a29, AutoModelForTokenClassification, _a30, AutoModelForSeq2SeqLM, _a31, AutoModelForSpeechSeq2Seq, _a32, AutoModelForTextToSpectrogram, _a33, AutoModelForTextToWaveform, _a34, AutoModelForCausalLM, _a35, AutoModelForMaskedLM, _a36, AutoModelForQuestionAnswering, _a37, AutoModelForVision2Seq, _a38, AutoModelForImageClassification, _a39, AutoModelForImageSegmentation, _a40, AutoModelForSemanticSegmentation, _a41, AutoModelForUniversalSegmentation, _a42, AutoModelForObjectDetection, _a43, AutoModelForZeroShotObjectDetection, _a44, AutoModelForMaskGeneration, _a45, AutoModelForCTC, _a46, AutoModelForAudioClassification, _a47, AutoModelForXVector, _a48, AutoModelForAudioFrameClassification, _a49, AutoModelForDocumentQuestionAnswering, _a50, AutoModelForImageMatting, _a51, AutoModelForImageToImage, _a52, AutoModelForDepthEstimation, _a53, AutoModelForNormalEstimation, _a54, AutoModelForPoseEstimation, _a55, AutoModelForImageFeatureExtraction, _a56, AutoModelForImageTextToText, _a57, AutoModelForAudioTextToText, Seq2SeqLMOutput, SequenceClassifierOutput, XVectorOutput, TokenClassifierOutput, MaskedLMOutput, QuestionAnsweringModelOutput, CausalLMOutput, CausalLMOutputWithPast, ImageMattingOutput, VitsModelOutput, Pipeline, TextClassificationPipeline, TokenClassificationPipeline, QuestionAnsweringPipeline, FillMaskPipeline, Text2TextGenerationPipeline, SummarizationPipeline, TranslationPipeline, TextGenerationPipeline, ZeroShotClassificationPipeline, AudioClassificationPipeline, ZeroShotAudioClassificationPipeline, AutomaticSpeechRecognitionPipeline, TextToAudioPipeline, ImageToTextPipeline, ImageClassificationPipeline, ImageSegmentationPipeline, BackgroundRemovalPipeline, ZeroShotImageClassificationPipeline, ObjectDetectionPipeline, ZeroShotObjectDetectionPipeline, DocumentQuestionAnsweringPipeline, ImageToImagePipeline, DepthEstimationPipeline, FeatureExtractionPipeline, ImageFeatureExtractionPipeline, SUPPORTED_TASKS, TASK_ALIASES, RawVideoFrame, RawVideo, BaseStreamer, stdout_write, TextStreamer, WhisperTextStreamer;
+var ONNX_WEB, import_meta, __defProp2, __export2, emptyObj, node_fs_default, emptyObj2, node_path_default, emptyObj3, node_url_default, VERSION, IS_PROCESS_AVAILABLE, IS_NODE_ENV, IS_FS_AVAILABLE, IS_PATH_AVAILABLE, IS_DENO_RUNTIME, IS_BUN_RUNTIME, IS_BROWSER_ENV, IS_WEBWORKER_ENV, IS_WEB_CACHE_AVAILABLE, IS_WEBGPU_AVAILABLE, IS_WEBNN_AVAILABLE, isSafari2, IS_SAFARI, apis, RUNNING_LOCALLY, dirname__, DEFAULT_CACHE_DIR, DEFAULT_LOCAL_MODEL_PATH, localModelPath, env2, Callable, CONTENT_TYPE_MAP, FileResponse, FileCache, ERROR_MAPPING, MAX_EXTERNAL_DATA_CHUNKS, REPO_ID_REGEX, P2FFT, NP2FFT, FFT, uint16_to_float32, onnxruntime_node_exports, noop, emptyObj4, onnxruntime_node_default, Readable, pipeline, createWriteStream, createReadStream, DEVICE_TO_EXECUTION_PROVIDER_MAPPING, LOG_LEVELS, DEFAULT_LOG_LEVEL, supportedDevices, defaultDevices, ONNX, ORT_SYMBOL, InferenceSession2, IS_WEB_ENV, webInitChain, wasmLoadPromise, webInferenceChain, ONNX_ENV, wrap, _a, TensorOpRegistry, DataTypeMap, Tensor22, arrayToIndexTensor, PriorityQueue, CharTrie, CharTrieNode, TokenLattice, TokenLatticeNode, DictionarySplitter, LRUCache, TOKEN_TYPES, Token, ORDERED_MAPPING_TABLE, ESCAPE_CHARACTERS, Statement, Program, If, For, Break, Continue, SetStatement, Macro, Comment, Expression, MemberExpression, CallExpression, Identifier, Literal, IntegerLiteral, FloatLiteral, StringLiteral, ArrayLiteral, TupleLiteral, ObjectLiteral, BinaryExpression, FilterExpression, FilterStatement, SelectExpression, TestExpression, UnaryExpression, SliceExpression, KeywordArgumentExpression, SpreadExpression, CallStatement, Ternary, BreakControl, ContinueControl, RuntimeValue, IntegerValue, FloatValue, StringValue, BooleanValue, ObjectValue, KeywordArgumentsValue, ArrayValue, TupleValue, FunctionValue, NullValue, UndefinedValue, Environment, Interpreter, NEWLINE, OPEN_STATEMENT, CLOSE_STATEMENT, Template, WHISPER_LANGUAGES, WHISPER_LANGUAGE_MAPPING, WHISPER_TO_LANGUAGE_CODE_MAPPING, PUNCTUATION_REGEX, PUNCTUATION_ONLY_REGEX, BLOOM_SPLIT_CHARS, PROBLEMATIC_REGEX_MAP, AddedToken, TokenizerModel, WordPieceTokenizer, Unigram, BYTES_TO_UNICODE, UNICODE_TO_BYTES, BPE, LegacyTokenizerModel, Normalizer, Replace, UnicodeNormalizer, NFC, NFD, NFKC, NFKD, StripNormalizer, StripAccents, Lowercase, Prepend, NormalizerSequence, BertNormalizer, PreTokenizer, BertPreTokenizer, ByteLevelPreTokenizer, SplitPreTokenizer, PunctuationPreTokenizer, DigitsPreTokenizer, PostProcessor, BertProcessing, RobertaProcessing, TemplateProcessing, ByteLevelPostProcessor, PostProcessorSequence, Decoder, ReplaceDecoder, ByteFallback, FuseDecoder, StripDecoder, WordPieceDecoder, ByteLevelDecoder, CTCDecoder, DecoderSequence, BPEDecoder, VitsDecoder, MetaspacePreTokenizer, MetaspaceDecoder, Precompiled, PreTokenizerSequence, WhitespacePreTokenizer, WhitespaceSplit, ReplacePreTokenizer, FixedLengthPreTokenizer, SPECIAL_TOKEN_ATTRIBUTES, PreTrainedTokenizer, TokenizersBackend, BertTokenizer, AlbertTokenizer, MobileBertTokenizer, SqueezeBertTokenizer, DebertaTokenizer, DebertaV2Tokenizer, HerbertTokenizer, ConvBertTokenizer, RoFormerTokenizer, DistilBertTokenizer, CamembertTokenizer, XLMTokenizer, ElectraTokenizer, T5Tokenizer, GPT2Tokenizer, BartTokenizer, MBartTokenizer, MBart50Tokenizer, RobertaTokenizer, BloomTokenizer, SPIECE_UNDERLINE, LlamaTokenizer, CodeLlamaTokenizer, XLMRobertaTokenizer, MPNetTokenizer, FalconTokenizer, GPTNeoXTokenizer, EsmTokenizer, Qwen2Tokenizer, GemmaTokenizer, Grok1Tokenizer, NllbTokenizer, M2M100Tokenizer, WhisperTokenizer, CodeGenTokenizer, CLIPTokenizer, SiglipTokenizer, MarianTokenizer, Wav2Vec2CTCTokenizer, BlenderbotTokenizer, BlenderbotSmallTokenizer, SpeechT5Tokenizer, NougatTokenizer, VitsTokenizer, CohereTokenizer, MgpstrTokenizer, Ernie4_5_Tokenizer, _a2, AutoTokenizer, GITHUB_ISSUE_URL, FEATURE_EXTRACTOR_NAME, IMAGE_PROCESSOR_NAME, PROCESSOR_NAME, CHAT_TEMPLATE_NAME, _a3, Processor, processors_exports, FeatureExtractor, feature_extractors_exports, noop2, Readable2, noop3, pipeline2, HERTZ_TO_MEL_MAPPING, MEL_TO_HERTZ_MAPPING, RawAudio, ASTFeatureExtractor, EncodecFeatureExtractor, ChatterboxFeatureExtractor, ClapFeatureExtractor, DacFeatureExtractor, Gemma3nAudioFeatureExtractor, MoonshineFeatureExtractor, EPSILON, ParakeetFeatureExtractor, PyAnnoteFeatureExtractor, SeamlessM4TFeatureExtractor, SnacFeatureExtractor, SpeechT5FeatureExtractor, Wav2Vec2FeatureExtractor, WeSpeakerFeatureExtractor, WhisperFeatureExtractor, emptyObj5, sharp_default, createCanvasFunction, ImageDataClass, loadImageFunction, IS_BROWSER_OR_WEBWORKER, RESAMPLING_MAPPING, CONTENT_TYPE_MAP2, RawImage, load_image, ImageProcessor, AutoFeatureExtractor, _a4, ChatterboxProcessor, image_processors_exports, BeitFeatureExtractor, BitImageProcessor, ChineseCLIPFeatureExtractor, CLIPImageProcessor, CLIPFeatureExtractor, ConvNextImageProcessor, ConvNextFeatureExtractor, DeiTImageProcessor, DeiTFeatureExtractor, DetrImageProcessor, DetrFeatureExtractor, DINOv3ViTImageProcessor, DonutImageProcessor, DonutFeatureExtractor, DPTImageProcessor, DPTFeatureExtractor, EfficientNetImageProcessor, GLPNFeatureExtractor, GroundingDinoImageProcessor, Idefics3ImageProcessor, VLMImageProcessor, JinaCLIPImageProcessor, LlavaOnevisionImageProcessor, MaskFormerImageProcessor, MaskFormerFeatureExtractor, Mask2FormerImageProcessor, MobileNetV1ImageProcessor, MobileNetV1FeatureExtractor, MobileNetV2ImageProcessor, MobileNetV2FeatureExtractor, MobileNetV3ImageProcessor, MobileNetV3FeatureExtractor, MobileNetV4ImageProcessor, MobileNetV4FeatureExtractor, MobileViTImageProcessor, MobileViTFeatureExtractor, NougatImageProcessor, OwlViTImageProcessor, OwlViTFeatureExtractor, Owlv2ImageProcessor, IMAGE_SIZE, SLICE_AXES, ceil, floor, sqrt, Phi3VImageProcessor, PvtImageProcessor, Qwen2VLImageProcessor, RTDetrImageProcessor, SamImageProcessor, SegformerImageProcessor, SegformerFeatureExtractor, SiglipImageProcessor, Swin2SRImageProcessor, ViTImageProcessor, ViTFeatureExtractor, VitMatteImageProcessor, VitPoseImageProcessor, YolosImageProcessor, YolosFeatureExtractor, AutoImageProcessor, _a5, Florence2Processor, _a6, Gemma3nProcessor, _a7, GroundingDinoProcessor, _a8, Idefics3Processor, _a9, VLChatProcessor, _a10, JinaCLIPProcessor, _a11, LlavaProcessor, DECODE_TYPE_MAPPING, _a12, MgpstrProcessor, _a13, MoonshineProcessor, _a14, OwlViTProcessor, IMAGE_TOKEN, IMAGE_TOKEN_PATTERN, _a15, Phi3VProcessor, IMAGE_TOKEN2, _a16, PaliGemmaProcessor, _a17, PyAnnoteProcessor, _a18, Qwen2VLProcessor, _a19, SamProcessor, Sam2Processor, Sam2VideoProcessor, _a20, SpeechT5Processor, _a21, UltravoxProcessor, AUDIO_TOKEN, BEGIN_AUDIO_TOKEN, NUM_AUDIO_TOKENS, _a22, VoxtralProcessor, _a23, Wav2Vec2Processor, _a24, Wav2Vec2ProcessorWithLM, _a25, WhisperProcessor, AutoProcessor, PretrainedConfig, AutoConfig, DEVICE_TYPES, isWebGpuFp16Supported, DATA_TYPES, DEFAULT_DEVICE_DTYPE_MAPPING, DEFAULT_DTYPE_SUFFIX_MAPPING, LogitsProcessor, LogitsWarper, LogitsProcessorList, ForcedBOSTokenLogitsProcessor, ForcedEOSTokenLogitsProcessor, SuppressTokensAtBeginLogitsProcessor, WhisperTimeStampLogitsProcessor, NoRepeatNGramLogitsProcessor, RepetitionPenaltyLogitsProcessor, MinLengthLogitsProcessor, MinNewTokensLengthLogitsProcessor, NoBadWordsLogitsProcessor, ClassifierFreeGuidanceLogitsProcessor, TemperatureLogitsWarper, TopPLogitsWarper, TopKLogitsWarper, GenerationConfig, StoppingCriteria, StoppingCriteriaList, MaxLengthCriteria, EosTokenCriteria, InterruptableStoppingCriteria, LogitsSampler, GreedySampler, MultinomialSampler, BeamSearchSampler, WhisperGenerationConfig, MODEL_TYPES, MODEL_TYPE_MAPPING, MODEL_NAME_TO_CLASS_MAPPING, MODEL_CLASS_TO_NAME_MAPPING, PreTrainedModel, ModelOutput, BaseModelOutput, BertPreTrainedModel, BertModel, BertForMaskedLM, BertForSequenceClassification, BertForTokenClassification, BertForQuestionAnswering, NeoBertPreTrainedModel, NeoBertModel, NeoBertForMaskedLM, NeoBertForSequenceClassification, NeoBertForTokenClassification, NeoBertForQuestionAnswering, ModernBertPreTrainedModel, ModernBertModel, ModernBertForMaskedLM, ModernBertForSequenceClassification, ModernBertForTokenClassification, ModernBertDecoderPreTrainedModel, ModernBertDecoderModel, ModernBertDecoderForCausalLM, NomicBertPreTrainedModel, NomicBertModel, RoFormerPreTrainedModel, RoFormerModel, RoFormerForMaskedLM, RoFormerForSequenceClassification, RoFormerForTokenClassification, RoFormerForQuestionAnswering, ConvBertPreTrainedModel, ConvBertModel, ConvBertForMaskedLM, ConvBertForSequenceClassification, ConvBertForTokenClassification, ConvBertForQuestionAnswering, ElectraPreTrainedModel, ElectraModel, ElectraForMaskedLM, ElectraForSequenceClassification, ElectraForTokenClassification, ElectraForQuestionAnswering, CamembertPreTrainedModel, CamembertModel, CamembertForMaskedLM, CamembertForSequenceClassification, CamembertForTokenClassification, CamembertForQuestionAnswering, DebertaPreTrainedModel, DebertaModel, DebertaForMaskedLM, DebertaForSequenceClassification, DebertaForTokenClassification, DebertaForQuestionAnswering, DebertaV2PreTrainedModel, DebertaV2Model, DebertaV2ForMaskedLM, DebertaV2ForSequenceClassification, DebertaV2ForTokenClassification, DebertaV2ForQuestionAnswering, DistilBertPreTrainedModel, DistilBertModel, DistilBertForSequenceClassification, DistilBertForTokenClassification, DistilBertForQuestionAnswering, DistilBertForMaskedLM, EsmPreTrainedModel, EsmModel, EsmForMaskedLM, EsmForSequenceClassification, EsmForTokenClassification, MobileBertPreTrainedModel, MobileBertModel, MobileBertForMaskedLM, MobileBertForSequenceClassification, MobileBertForQuestionAnswering, MPNetPreTrainedModel, MPNetModel, MPNetForMaskedLM, MPNetForSequenceClassification, MPNetForTokenClassification, MPNetForQuestionAnswering, SqueezeBertPreTrainedModel, SqueezeBertModel, SqueezeBertForMaskedLM, SqueezeBertForSequenceClassification, SqueezeBertForQuestionAnswering, AlbertPreTrainedModel, AlbertModel, AlbertForSequenceClassification, AlbertForQuestionAnswering, AlbertForMaskedLM, T5PreTrainedModel, T5Model, T5ForConditionalGeneration, LongT5PreTrainedModel, LongT5Model, LongT5ForConditionalGeneration, MT5PreTrainedModel, MT5Model, MT5ForConditionalGeneration, BartPretrainedModel, BartModel, BartForConditionalGeneration, BartForSequenceClassification, MBartPreTrainedModel, MBartModel, MBartForConditionalGeneration, MBartForSequenceClassification, MBartForCausalLM, BlenderbotPreTrainedModel, BlenderbotModel, BlenderbotForConditionalGeneration, BlenderbotSmallPreTrainedModel, BlenderbotSmallModel, BlenderbotSmallForConditionalGeneration, RobertaPreTrainedModel, RobertaModel, RobertaForMaskedLM, RobertaForSequenceClassification, RobertaForTokenClassification, RobertaForQuestionAnswering, XLMPreTrainedModel, XLMModel, XLMWithLMHeadModel, XLMForSequenceClassification, XLMForTokenClassification, XLMForQuestionAnswering, XLMRobertaPreTrainedModel, XLMRobertaModel, XLMRobertaForMaskedLM, XLMRobertaForSequenceClassification, XLMRobertaForTokenClassification, XLMRobertaForQuestionAnswering, ASTPreTrainedModel, ASTModel, ASTForAudioClassification, WhisperPreTrainedModel, WhisperModel, WhisperForConditionalGeneration, LiteWhisperForConditionalGeneration, MoonshinePreTrainedModel, MoonshineModel, MoonshineForConditionalGeneration, VisionEncoderDecoderModel, LlavaPreTrainedModel, LlavaForConditionalGeneration, LlavaOnevisionForConditionalGeneration, Moondream1ForConditionalGeneration, Florence2PreTrainedModel, Florence2ForConditionalGeneration, PaliGemmaPreTrainedModel, PaliGemmaForConditionalGeneration, LlavaQwen2ForCausalLM, Gemma3nPreTrainedModel, Gemma3nForConditionalGeneration, Idefics3PreTrainedModel, Idefics3ForConditionalGeneration, SmolVLMForConditionalGeneration, Phi3VPreTrainedModel, Phi3VForCausalLM, CLIPPreTrainedModel, CLIPModel, CLIPTextModel, CLIPTextModelWithProjection, CLIPVisionModel, CLIPVisionModelWithProjection, SiglipPreTrainedModel, SiglipModel, SiglipTextModel, SiglipVisionModel, ChineseCLIPPreTrainedModel, ChineseCLIPModel, JinaCLIPPreTrainedModel, JinaCLIPModel, JinaCLIPTextModel, JinaCLIPVisionModel, CLIPSegPreTrainedModel, CLIPSegModel, CLIPSegForImageSegmentation, GPT2PreTrainedModel, GPT2Model, GPT2LMHeadModel, GptOssPreTrainedModel, GptOssModel, GptOssForCausalLM, JAISPreTrainedModel, JAISModel, JAISLMHeadModel, GPTNeoPreTrainedModel, GPTNeoModel, GPTNeoForCausalLM, GPTNeoXPreTrainedModel, GPTNeoXModel, GPTNeoXForCausalLM, GPTJPreTrainedModel, GPTJModel, GPTJForCausalLM, GPTBigCodePreTrainedModel, GPTBigCodeModel, GPTBigCodeForCausalLM, CodeGenPreTrainedModel, CodeGenModel, CodeGenForCausalLM, LlamaPreTrainedModel, LlamaModel, LlamaForCausalLM, Llama4PreTrainedModel, Llama4ForCausalLM, NanoChatPreTrainedModel, NanoChatModel, NanoChatForCausalLM, ApertusPreTrainedModel, ApertusModel, ApertusForCausalLM, ArceePreTrainedModel, ArceeModel, ArceeForCausalLM, Lfm2PreTrainedModel, Lfm2Model, Lfm2ForCausalLM, SmolLM3PreTrainedModel, SmolLM3Model, SmolLM3ForCausalLM, HeliumPreTrainedModel, HeliumModel, HeliumForCausalLM, GlmPreTrainedModel, GlmModel, GlmForCausalLM, ExaonePreTrainedModel, ExaoneModel, ExaoneForCausalLM, MobileLLMPreTrainedModel, MobileLLMModel, MobileLLMForCausalLM, OlmoPreTrainedModel, OlmoModel, OlmoForCausalLM, Olmo2PreTrainedModel, Olmo2Model, Olmo2ForCausalLM, Olmo3PreTrainedModel, Olmo3Model, Olmo3ForCausalLM, GranitePreTrainedModel, GraniteModel, GraniteForCausalLM, GraniteMoeHybridPreTrainedModel, GraniteMoeHybridModel, GraniteMoeHybridForCausalLM, CoherePreTrainedModel, CohereModel, CohereForCausalLM, GemmaPreTrainedModel, GemmaModel, GemmaForCausalLM, Gemma2PreTrainedModel, Gemma2Model, Gemma2ForCausalLM, VaultGemmaPreTrainedModel, VaultGemmaModel, VaultGemmaForCausalLM, Gemma3PreTrainedModel, Gemma3Model, Gemma3ForCausalLM, OpenELMPreTrainedModel, OpenELMModel, OpenELMForCausalLM, Qwen2PreTrainedModel, Qwen2Model, Qwen2ForCausalLM, Qwen3PreTrainedModel, Qwen3Model, Qwen3ForCausalLM, Qwen2VLPreTrainedModel, Qwen2VLForConditionalGeneration, PhiPreTrainedModel, PhiModel, PhiForCausalLM, Phi3PreTrainedModel, Phi3Model, Phi3ForCausalLM, BloomPreTrainedModel, BloomModel, BloomForCausalLM, MptPreTrainedModel, MptModel, MptForCausalLM, OPTPreTrainedModel, OPTModel, OPTForCausalLM, ViTPreTrainedModel, ViTModel, ViTForImageClassification, IJepaPreTrainedModel, IJepaModel, IJepaForImageClassification, VitPosePreTrainedModel, VitPoseForPoseEstimation, PvtPreTrainedModel, PvtModel, PvtForImageClassification, ViTMAEPreTrainedModel, ViTMAEModel, ViTMSNPreTrainedModel, ViTMSNModel, ViTMSNForImageClassification, GroupViTPreTrainedModel, GroupViTModel, FastViTPreTrainedModel, FastViTModel, FastViTForImageClassification, VitMattePreTrainedModel, VitMatteForImageMatting, MobileViTPreTrainedModel, MobileViTModel, MobileViTForImageClassification, MobileViTV2PreTrainedModel, MobileViTV2Model, MobileViTV2ForImageClassification, OwlViTPreTrainedModel, OwlViTModel, OwlViTForObjectDetection, Owlv2PreTrainedModel, Owlv2Model, Owlv2ForObjectDetection, BeitPreTrainedModel, BeitModel, BeitForImageClassification, DetrPreTrainedModel, DetrModel, DetrForObjectDetection, DetrForSegmentation, DetrObjectDetectionOutput, DetrSegmentationOutput, RTDetrPreTrainedModel, RTDetrModel, RTDetrForObjectDetection, RTDetrObjectDetectionOutput, RTDetrV2PreTrainedModel, RTDetrV2Model, RTDetrV2ForObjectDetection, RTDetrV2ObjectDetectionOutput, RFDetrPreTrainedModel, RFDetrModel, RFDetrForObjectDetection, RFDetrObjectDetectionOutput, DFinePreTrainedModel, DFineModel, DFineForObjectDetection, TableTransformerPreTrainedModel, TableTransformerModel, TableTransformerForObjectDetection, TableTransformerObjectDetectionOutput, DeiTPreTrainedModel, DeiTModel, DeiTForImageClassification, HieraPreTrainedModel, HieraModel, HieraForImageClassification, ResNetPreTrainedModel, ResNetModel, ResNetForImageClassification, SwinPreTrainedModel, SwinModel, SwinForImageClassification, SwinForSemanticSegmentation, Swin2SRPreTrainedModel, Swin2SRModel, Swin2SRForImageSuperResolution, DPTPreTrainedModel, DPTModel, DPTForDepthEstimation, DepthAnythingPreTrainedModel, DepthAnythingForDepthEstimation, SapiensPreTrainedModel, SapiensForSemanticSegmentation, SapiensForDepthEstimation, SapiensForNormalEstimation, DepthProPreTrainedModel, DepthProForDepthEstimation, Metric3DPreTrainedModel, Metric3DForDepthEstimation, Metric3Dv2PreTrainedModel, Metric3Dv2ForDepthEstimation, MaskFormerPreTrainedModel, MaskFormerModel, MaskFormerForInstanceSegmentation, GLPNPreTrainedModel, GLPNModel, GLPNForDepthEstimation, DonutSwinPreTrainedModel, DonutSwinModel, ConvNextPreTrainedModel, ConvNextModel, ConvNextForImageClassification, ConvNextV2PreTrainedModel, ConvNextV2Model, ConvNextV2ForImageClassification, Dinov2PreTrainedModel, Dinov2Model, Dinov2ForImageClassification, Dinov2WithRegistersPreTrainedModel, Dinov2WithRegistersModel, Dinov2WithRegistersForImageClassification, DINOv3ViTPreTrainedModel, DINOv3ViTModel, DINOv3ConvNextPreTrainedModel, DINOv3ConvNextModel, GroundingDinoPreTrainedModel, GroundingDinoForObjectDetection, YolosPreTrainedModel, YolosModel, YolosForObjectDetection, YolosObjectDetectionOutput, SamPreTrainedModel, SamModel, SamImageSegmentationOutput, Sam2ImageSegmentationOutput, Sam2PreTrainedModel, Sam2Model, EdgeTamModel, Sam3TrackerModel, MarianPreTrainedModel, MarianModel, MarianMTModel, M2M100PreTrainedModel, M2M100Model, M2M100ForConditionalGeneration, Wav2Vec2PreTrainedModel, Wav2Vec2Model, Wav2Vec2ForCTC, Wav2Vec2ForSequenceClassification, Wav2Vec2ForAudioFrameClassification, ParakeetPreTrainedModel, ParakeetForCTC, PyAnnotePreTrainedModel, PyAnnoteModel, PyAnnoteForAudioFrameClassification, WeSpeakerResNetPreTrainedModel, WeSpeakerResNetModel, UniSpeechPreTrainedModel, UniSpeechModel, UniSpeechForCTC, UniSpeechForSequenceClassification, UniSpeechSatPreTrainedModel, UniSpeechSatModel, UniSpeechSatForCTC, UniSpeechSatForSequenceClassification, UniSpeechSatForAudioFrameClassification, Wav2Vec2BertPreTrainedModel, Wav2Vec2BertModel, Wav2Vec2BertForCTC, Wav2Vec2BertForSequenceClassification, HubertPreTrainedModel, HubertModel, HubertForCTC, HubertForSequenceClassification, WavLMPreTrainedModel, WavLMModel, WavLMForCTC, WavLMForSequenceClassification, WavLMForXVector, WavLMForAudioFrameClassification, StyleTextToSpeech2PreTrainedModel, StyleTextToSpeech2Model, SpeechT5PreTrainedModel, SpeechT5Model, SpeechT5ForSpeechToText, SpeechT5ForTextToSpeech, SpeechT5HifiGan, SupertonicPreTrainedModel, SupertonicForConditionalGeneration, TrOCRPreTrainedModel, TrOCRForCausalLM, MistralPreTrainedModel, MistralModel, MistralForCausalLM, Ernie4_5_PretrainedModel, Ernie4_5_Model, Ernie4_5_ForCausalLM, Starcoder2PreTrainedModel, Starcoder2Model, Starcoder2ForCausalLM, FalconPreTrainedModel, FalconModel, FalconForCausalLM, ClapPreTrainedModel, ClapModel, ClapTextModelWithProjection, ClapAudioModelWithProjection, VitsPreTrainedModel, VitsModel, SegformerPreTrainedModel, SegformerModel, SegformerForImageClassification, SegformerForSemanticSegmentation, StableLmPreTrainedModel, StableLmModel, StableLmForCausalLM, EfficientNetPreTrainedModel, EfficientNetModel, EfficientNetForImageClassification, MusicgenPreTrainedModel, MusicgenModel, MusicgenForCausalLM, MusicgenForConditionalGeneration, MobileNetV1PreTrainedModel, MobileNetV1Model, MobileNetV1ForImageClassification, MobileNetV1ForSemanticSegmentation, MobileNetV2PreTrainedModel, MobileNetV2Model, MobileNetV2ForImageClassification, MobileNetV2ForSemanticSegmentation, MobileNetV3PreTrainedModel, MobileNetV3Model, MobileNetV3ForImageClassification, MobileNetV3ForSemanticSegmentation, MobileNetV4PreTrainedModel, MobileNetV4Model, MobileNetV4ForImageClassification, MobileNetV4ForSemanticSegmentation, DecisionTransformerPreTrainedModel, DecisionTransformerModel, MultiModalityPreTrainedModel, MultiModalityCausalLM, MgpstrModelOutput, MgpstrPreTrainedModel, MgpstrForSceneTextRecognition, PatchTSTPreTrainedModel, PatchTSTModel, PatchTSTForPrediction, PatchTSMixerPreTrainedModel, PatchTSMixerModel, PatchTSMixerForPrediction, UltravoxPreTrainedModel, UltravoxModel, VoxtralForConditionalGeneration, MimiPreTrainedModel, MimiEncoderOutput, MimiDecoderOutput, MimiModel, MimiEncoderModel, MimiDecoderModel, DacPreTrainedModel, DacEncoderOutput, DacDecoderOutput, DacModel, DacEncoderModel, DacDecoderModel, SnacPreTrainedModel, SnacModel, SnacEncoderModel, SnacDecoderModel, ChatterboxPreTrainedModel, ChatterboxModel, _a26, PretrainedMixin, MODEL_MAPPING_NAMES_ENCODER_ONLY, MODEL_MAPPING_NAMES_ENCODER_DECODER, MODEL_MAPPING_NAMES_AUTO_ENCODER, MODEL_MAPPING_NAMES_DECODER_ONLY, MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING_NAMES, MODEL_FOR_TEXT_TO_SPECTROGRAM_MAPPING_NAMES, MODEL_FOR_TEXT_TO_WAVEFORM_MAPPING_NAMES, MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES, MODEL_FOR_CAUSAL_LM_MAPPING_NAMES, MODEL_FOR_MULTIMODALITY_MAPPING_NAMES, MODEL_FOR_MASKED_LM_MAPPING_NAMES, MODEL_FOR_QUESTION_ANSWERING_MAPPING_NAMES, MODEL_FOR_VISION_2_SEQ_MAPPING_NAMES, MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES, MODEL_FOR_AUDIO_TEXT_TO_TEXT_MAPPING_NAMES, MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES, MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES, MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING_NAMES, MODEL_FOR_IMAGE_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_UNIVERSAL_SEGMENTATION_MAPPING_NAMES, MODEL_FOR_MASK_GENERATION_MAPPING_NAMES, MODEL_FOR_CTC_MAPPING_NAMES, MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_AUDIO_XVECTOR_MAPPING_NAMES, MODEL_FOR_AUDIO_FRAME_CLASSIFICATION_MAPPING_NAMES, MODEL_FOR_IMAGE_MATTING_MAPPING_NAMES, MODEL_FOR_TIME_SERIES_PREDICTION_MAPPING_NAMES, MODEL_FOR_IMAGE_TO_IMAGE_MAPPING_NAMES, MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES, MODEL_FOR_NORMAL_ESTIMATION_MAPPING_NAMES, MODEL_FOR_POSE_ESTIMATION_MAPPING_NAMES, MODEL_FOR_IMAGE_FEATURE_EXTRACTION_MAPPING_NAMES, MODEL_CLASS_TYPE_MAPPING, CUSTOM_MAPPING, CUSTOM_ARCHITECTURES, _a27, AutoModel, _a28, AutoModelForSequenceClassification, _a29, AutoModelForTokenClassification, _a30, AutoModelForSeq2SeqLM, _a31, AutoModelForSpeechSeq2Seq, _a32, AutoModelForTextToSpectrogram, _a33, AutoModelForTextToWaveform, _a34, AutoModelForCausalLM, _a35, AutoModelForMaskedLM, _a36, AutoModelForQuestionAnswering, _a37, AutoModelForVision2Seq, _a38, AutoModelForImageClassification, _a39, AutoModelForImageSegmentation, _a40, AutoModelForSemanticSegmentation, _a41, AutoModelForUniversalSegmentation, _a42, AutoModelForObjectDetection, _a43, AutoModelForZeroShotObjectDetection, _a44, AutoModelForMaskGeneration, _a45, AutoModelForCTC, _a46, AutoModelForAudioClassification, _a47, AutoModelForXVector, _a48, AutoModelForAudioFrameClassification, _a49, AutoModelForDocumentQuestionAnswering, _a50, AutoModelForImageMatting, _a51, AutoModelForImageToImage, _a52, AutoModelForDepthEstimation, _a53, AutoModelForNormalEstimation, _a54, AutoModelForPoseEstimation, _a55, AutoModelForImageFeatureExtraction, _a56, AutoModelForImageTextToText, _a57, AutoModelForAudioTextToText, Seq2SeqLMOutput, SequenceClassifierOutput, XVectorOutput, TokenClassifierOutput, MaskedLMOutput, QuestionAnsweringModelOutput, CausalLMOutput, CausalLMOutputWithPast, ImageMattingOutput, VitsModelOutput, Pipeline, TextClassificationPipeline, TokenClassificationPipeline, QuestionAnsweringPipeline, FillMaskPipeline, Text2TextGenerationPipeline, SummarizationPipeline, TranslationPipeline, TextGenerationPipeline, ZeroShotClassificationPipeline, AudioClassificationPipeline, ZeroShotAudioClassificationPipeline, AutomaticSpeechRecognitionPipeline, TextToAudioPipeline, ImageToTextPipeline, ImageClassificationPipeline, ImageSegmentationPipeline, BackgroundRemovalPipeline, ZeroShotImageClassificationPipeline, ObjectDetectionPipeline, ZeroShotObjectDetectionPipeline, DocumentQuestionAnsweringPipeline, ImageToImagePipeline, DepthEstimationPipeline, FeatureExtractionPipeline, ImageFeatureExtractionPipeline, SUPPORTED_TASKS, TASK_ALIASES, RawVideoFrame, RawVideo, BaseStreamer, stdout_write, TextStreamer, WhisperTextStreamer;
 var init_transformers_web = __esm({
   "node_modules/@huggingface/transformers/dist/transformers.web.js"() {
     "use strict";
@@ -6014,7 +6014,7 @@ var init_transformers_web = __esm({
     IS_WEB_CACHE_AVAILABLE = typeof self !== "undefined" && "caches" in self;
     IS_WEBGPU_AVAILABLE = IS_NODE_ENV || typeof navigator !== "undefined" && "gpu" in navigator;
     IS_WEBNN_AVAILABLE = typeof navigator !== "undefined" && "ml" in navigator;
-    isSafari = () => {
+    isSafari2 = () => {
       if (typeof navigator === "undefined") {
         return false;
       }
@@ -6024,7 +6024,7 @@ var init_transformers_web = __esm({
       const notOtherBrowser = !userAgent.match(/CriOS|FxiOS|EdgiOS|OPiOS|mercury|brave/i) && !userAgent.includes("Chrome") && !userAgent.includes("Android");
       return isAppleVendor && notOtherBrowser;
     };
-    IS_SAFARI = isSafari();
+    IS_SAFARI = isSafari2();
     apis = Object.freeze({
       /** Whether we are running in a browser environment (and not a web worker) */
       IS_BROWSER_ENV,
@@ -26505,6 +26505,7 @@ ${boi_token}${image_tokens_expanded}${eoi_token}
 // src/index.ts
 var index_exports = {};
 __export(index_exports, {
+  ARKIT_BLENDSHAPES: () => ARKIT_BLENDSHAPES,
   AgentCoreAdapter: () => AgentCoreAdapter,
   AnimationGraph: () => AnimationGraph,
   AudioChunkCoalescer: () => AudioChunkCoalescer,
@@ -26540,6 +26541,8 @@ __export(index_exports, {
   SileroVADWorker: () => SileroVADWorker,
   SyncedAudioPipeline: () => SyncedAudioPipeline,
   TenantManager: () => TenantManager,
+  WAV2ARKIT_BLENDSHAPES: () => WAV2ARKIT_BLENDSHAPES,
+  Wav2ArkitCpuInference: () => Wav2ArkitCpuInference,
   Wav2Vec2Inference: () => Wav2Vec2Inference,
   WhisperInference: () => WhisperInference,
   blendEmotions: () => blendEmotions,
@@ -26551,6 +26554,7 @@ __export(index_exports, {
   configureLogging: () => configureLogging,
   configureTelemetry: () => configureTelemetry,
   createEmotionVector: () => createEmotionVector,
+  createLipSync: () => createLipSync,
   createLogger: () => createLogger,
   createSessionWithFallback: () => createSessionWithFallback,
   createSileroVAD: () => createSileroVAD,
@@ -26575,6 +26579,7 @@ __export(index_exports, {
   isIOSSafari: () => isIOSSafari,
   isMobile: () => isMobile,
   isOnnxRuntimeLoaded: () => isOnnxRuntimeLoaded,
+  isSafari: () => isSafari,
   isSpeechRecognitionAvailable: () => isSpeechRecognitionAvailable,
   isWebGPUAvailable: () => isWebGPUAvailable,
   lerpEmotion: () => lerpEmotion,
@@ -26583,15 +26588,18 @@ __export(index_exports, {
   nukeBrowserCaches: () => nukeBrowserCaches,
   parseHuggingFaceUrl: () => parseHuggingFaceUrl,
   preloadModels: () => preloadModels,
+  remapWav2ArkitToLam: () => remapWav2ArkitToLam,
   resetLoggingConfig: () => resetLoggingConfig,
   resolveBackend: () => resolveBackend,
   scanForInvalidCaches: () => scanForInvalidCaches,
   setLogLevel: () => setLogLevel,
   setLoggingEnabled: () => setLoggingEnabled,
   shouldEnableWasmProxy: () => shouldEnableWasmProxy,
+  shouldUseCpuLipSync: () => shouldUseCpuLipSync,
   shouldUseNativeASR: () => shouldUseNativeASR,
   shouldUseServerLipSync: () => shouldUseServerLipSync,
   supportsVADWorker: () => supportsVADWorker,
+  symmetrizeBlendshapes: () => symmetrizeBlendshapes,
   validateCachedResponse: () => validateCachedResponse
 });
 module.exports = __toCommonJS(index_exports);
@@ -26832,6 +26840,19 @@ var AudioScheduler = class {
   async initialize() {
     console.log("[AudioScheduler] Ready for lazy initialization");
   }
+  /**
+   * Eagerly create and warm up the AudioContext
+   *
+   * Call this when a playback session starts (e.g., when AI response begins).
+   * The AudioContext needs time to initialize the audio hardware — on Windows
+   * this can take 50-100ms. By warming up early (before audio data arrives),
+   * the context is fully ready when schedule() is first called.
+   *
+   * Must be called after a user gesture (click/tap) for autoplay policy.
+   */
+  async warmup() {
+    await this.ensureContext();
+  }
   /**
    * Ensure AudioContext is created and ready
    * Called lazily on first schedule() - requires user gesture
@@ -26862,7 +26883,7 @@ var AudioScheduler = class {
     const ctx = await this.ensureContext();
     const channels = this.options.channels ?? 1;
     if (!this.isPlaying) {
-      this.nextPlayTime = ctx.currentTime;
+      this.nextPlayTime = ctx.currentTime + 0.05;
       this.isPlaying = true;
     }
     const audioBuffer = ctx.createBuffer(channels, audioData.length, ctx.sampleRate);
@@ -26936,8 +26957,19 @@ var AudioScheduler = class {
   }
   /**
    * Reset scheduler state for new playback session
+   * Stops any orphaned sources that weren't cleaned up by cancelAll()
    */
   reset() {
+    if (this.context) {
+      const now = this.context.currentTime;
+      for (const { source, gainNode } of this.scheduledSources) {
+        try {
+          gainNode.gain.setValueAtTime(0, now);
+          source.stop(now);
+        } catch {
+        }
+      }
+    }
     this.nextPlayTime = 0;
     this.isPlaying = false;
     this.scheduledSources = [];
@@ -27065,7 +27097,7 @@ var LAMPipeline = class {
     newBuffer.set(this.buffer, 0);
     newBuffer.set(samples, this.buffer.length);
     this.buffer = newBuffer;
-    if (this.buffer.length >= this.REQUIRED_SAMPLES) {
+    while (this.buffer.length >= this.REQUIRED_SAMPLES) {
       await this.processBuffer(lam);
     }
   }
@@ -27218,12 +27250,20 @@ var LAMPipeline = class {
 };
 
 // src/audio/SyncedAudioPipeline.ts
+function pcm16ToFloat32(buffer) {
+  const byteLen = buffer.byteLength & ~1;
+  const int16 = byteLen === buffer.byteLength ? new Int16Array(buffer) : new Int16Array(buffer, 0, byteLen / 2);
+  const float32 = new Float32Array(int16.length);
+  for (let i = 0; i < int16.length; i++) {
+    float32[i] = int16[i] / 32768;
+  }
+  return float32;
+}
 var SyncedAudioPipeline = class extends EventEmitter {
   constructor(options) {
     super();
     this.options = options;
-    this.waitingForFirstLAM = false;
-    this.bufferedChunks = [];
+    this.playbackStarted = false;
     this.monitorInterval = null;
     this.frameAnimationId = null;
     const sampleRate = options.sampleRate ?? 16e3;
@@ -27234,11 +27274,6 @@ var SyncedAudioPipeline = class extends EventEmitter {
     });
     this.lamPipeline = new LAMPipeline({
       sampleRate,
-      onInference: (frameCount) => {
-        if (this.waitingForFirstLAM) {
-          this.onFirstLAMComplete();
-        }
-      },
       onError: (error) => {
         this.emit("error", error);
       }
@@ -27254,25 +27289,24 @@ var SyncedAudioPipeline = class extends EventEmitter {
    * Start a new playback session
    *
    * Resets all state and prepares for incoming audio chunks.
-   * Enables wait-for-first-LAM synchronization.
+   * Audio will be scheduled immediately as chunks arrive (no buffering).
    */
   start() {
+    this.stopMonitoring();
     this.scheduler.reset();
     this.coalescer.reset();
     this.lamPipeline.reset();
-    this.bufferedChunks = [];
-    this.waitingForFirstLAM = true;
+    this.playbackStarted = false;
+    this.scheduler.warmup();
     this.startFrameLoop();
     this.startMonitoring();
   }
   /**
    * Receive audio chunk from network
    *
-   * Implements wait-for-first-LAM pattern:
-   * - Chunks are coalesced into optimal buffers
-   * - Buffers are sent to LAM for processing
-   * - Audio scheduling waits until first LAM completes
-   * - Then all buffered audio is scheduled together with LAM frames
+   * Audio-first design: schedules audio immediately, LAM runs in background.
+   * This prevents LAM inference (50-300ms) from blocking audio scheduling,
+   * which caused audible stuttering with continuous audio streams.
    *
    * @param chunk - Uint8Array containing Int16 PCM audio
    */
@@ -27281,51 +27315,15 @@ var SyncedAudioPipeline = class extends EventEmitter {
     if (!combined) {
       return;
     }
-    const int16 = new Int16Array(combined);
-    const float32 = new Float32Array(int16.length);
-    for (let i = 0; i < int16.length; i++) {
-      float32[i] = int16[i] / 32768;
-    }
-    if (this.waitingForFirstLAM) {
-      this.bufferedChunks.push(combined);
-      const estimatedTime = this.scheduler.getCurrentTime();
-      await this.lamPipeline.push(float32, estimatedTime, this.options.lam);
-    } else {
-      const scheduleTime = await this.scheduler.schedule(float32);
-      await this.lamPipeline.push(float32, scheduleTime, this.options.lam);
-    }
-  }
-  /**
-   * Handle first LAM inference completion
-   *
-   * This is the critical synchronization point:
-   * - LAM frames are now ready in the queue
-   * - Schedule all buffered audio chunks
-   * - Adjust LAM frame timestamps to match actual schedule time
-   * - Audio and LAM start playing together, perfectly synchronized
-   */
-  async onFirstLAMComplete() {
-    this.waitingForFirstLAM = false;
-    const beforeSchedule = this.scheduler.getCurrentTime();
-    let actualStartTime = beforeSchedule;
-    for (let i = 0; i < this.bufferedChunks.length; i++) {
-      const buffer = this.bufferedChunks[i];
-      const int16 = new Int16Array(buffer);
-      const float32 = new Float32Array(int16.length);
-      for (let j = 0; j < int16.length; j++) {
-        float32[j] = int16[j] / 32768;
-      }
-      const scheduleTime = await this.scheduler.schedule(float32);
-      if (i === 0) {
-        actualStartTime = scheduleTime;
-      }
-    }
-    const timeOffset = actualStartTime - beforeSchedule;
-    if (timeOffset !== 0) {
-      this.lamPipeline.adjustTimestamps(timeOffset);
+    const float32 = pcm16ToFloat32(combined);
+    const scheduleTime = await this.scheduler.schedule(float32);
+    if (!this.playbackStarted) {
+      this.playbackStarted = true;
+      this.emit("playback_start", scheduleTime);
     }
-    this.bufferedChunks = [];
-    this.emit("playback_start", actualStartTime);
+    this.lamPipeline.push(float32, scheduleTime, this.options.lam).catch((err) => {
+      this.emit("error", err);
+    });
   }
   /**
    * End of audio stream
@@ -27357,10 +27355,9 @@ var SyncedAudioPipeline = class extends EventEmitter {
   async stop(fadeOutMs = 50) {
     this.stopMonitoring();
     await this.scheduler.cancelAll(fadeOutMs);
-    this.bufferedChunks = [];
     this.coalescer.reset();
     this.lamPipeline.reset();
-    this.waitingForFirstLAM = false;
+    this.playbackStarted = false;
     this.emit("playback_complete", void 0);
   }
   /**
@@ -27417,8 +27414,7 @@ var SyncedAudioPipeline = class extends EventEmitter {
    */
   getState() {
     return {
-      waitingForFirstLAM: this.waitingForFirstLAM,
-      bufferedChunks: this.bufferedChunks.length,
+      playbackStarted: this.playbackStarted,
       coalescerFill: this.coalescer.fillLevel,
       lamFill: this.lamPipeline.fillLevel,
       queuedFrames: this.lamPipeline.queuedFrameCount,
@@ -27434,7 +27430,6 @@ var SyncedAudioPipeline = class extends EventEmitter {
     this.scheduler.dispose();
     this.coalescer.reset();
     this.lamPipeline.reset();
-    this.bufferedChunks = [];
   }
 };
 
@@ -28876,12 +28871,12 @@ var Logger = class _Logger {
 };
 var loggerCache = /* @__PURE__ */ new Map();
 function createLogger(module2) {
-  let logger12 = loggerCache.get(module2);
-  if (!logger12) {
-    logger12 = new Logger(module2);
-    loggerCache.set(module2, logger12);
+  let logger13 = loggerCache.get(module2);
+  if (!logger13) {
+    logger13 = new Logger(module2);
+    loggerCache.set(module2, logger13);
   }
-  return logger12;
+  return logger13;
 }
 var noopLogger = {
   module: "noop",
@@ -28925,7 +28920,7 @@ function hasWebGPUApi() {
   return "gpu" in navigator && navigator.gpu !== void 0;
 }
 function getRecommendedBackend() {
-  if (isIOS()) {
+  if (isSafari() || isIOS()) {
     return "wasm";
   }
   return "webgpu";
@@ -28969,6 +28964,14 @@ function shouldEnableWasmProxy() {
   }
   return true;
 }
+function isSafari() {
+  if (typeof navigator === "undefined") return false;
+  const ua = navigator.userAgent.toLowerCase();
+  return /safari/.test(ua) && !/chrome|crios|fxios|chromium|edg/.test(ua);
+}
+function shouldUseCpuLipSync() {
+  return isSafari();
+}
 function isSpeechRecognitionAvailable() {
   if (typeof window === "undefined") return false;
   return "SpeechRecognition" in window || "webkitSpeechRecognition" in window;
@@ -29013,13 +29016,13 @@ async function isWebGPUAvailable() {
     return false;
   }
 }
-function configureWasm(ort2) {
-  ort2.env.wasm.wasmPaths = WASM_CDN_PATH;
+function configureWasm(ort) {
+  ort.env.wasm.wasmPaths = WASM_CDN_PATH;
   const numThreads = getOptimalWasmThreads();
   const enableProxy = shouldEnableWasmProxy();
-  ort2.env.wasm.numThreads = numThreads;
-  ort2.env.wasm.simd = true;
-  ort2.env.wasm.proxy = enableProxy;
+  ort.env.wasm.numThreads = numThreads;
+  ort.env.wasm.simd = true;
+  ort.env.wasm.proxy = enableProxy;
   logger.info("WASM configured", {
     numThreads,
     simd: true,
@@ -29067,8 +29070,8 @@ async function getOnnxRuntimeForPreference(preference = "auto") {
     webgpuAvailable,
     resolvedBackend: backend
   });
-  const ort2 = await getOnnxRuntime(backend);
-  return { ort: ort2, backend };
+  const ort = await getOnnxRuntime(backend);
+  return { ort, backend };
 }
 function getSessionOptions(backend) {
   if (backend === "webgpu") {
@@ -29089,12 +29092,12 @@ function getSessionOptions(backend) {
   };
 }
 async function createSessionWithFallback(modelBuffer, preferredBackend) {
-  const ort2 = await getOnnxRuntime(preferredBackend);
+  const ort = await getOnnxRuntime(preferredBackend);
   const modelData = new Uint8Array(modelBuffer);
   if (preferredBackend === "webgpu") {
     try {
       const options2 = getSessionOptions("webgpu");
-      const session2 = await ort2.InferenceSession.create(modelData, options2);
+      const session2 = await ort.InferenceSession.create(modelData, options2);
       logger.info("Session created with WebGPU backend");
       return { session: session2, backend: "webgpu" };
     } catch (err) {
@@ -29104,7 +29107,7 @@ async function createSessionWithFallback(modelBuffer, preferredBackend) {
     }
   }
   const options = getSessionOptions("wasm");
-  const session = await ort2.InferenceSession.create(modelData, options);
+  const session = await ort.InferenceSession.create(modelData, options);
   logger.info("Session created with WASM backend");
   return { session, backend: "wasm" };
 }
@@ -29115,8 +29118,7 @@ function isOnnxRuntimeLoaded() {
   return ortInstance !== null;
 }
 
-// src/inference/Wav2Vec2Inference.ts
-var logger2 = createLogger("Wav2Vec2");
+// src/inference/blendshapeUtils.ts
 var LAM_BLENDSHAPES = [
   "browDownLeft",
   "browDownRight",
@@ -29171,40 +29173,7 @@ var LAM_BLENDSHAPES = [
   "noseSneerRight",
   "tongueOut"
 ];
-var CTC_VOCAB = [
-  "<pad>",
-  "<s>",
-  "</s>",
-  "<unk>",
-  "|",
-  "E",
-  "T",
-  "A",
-  "O",
-  "N",
-  "I",
-  "H",
-  "S",
-  "R",
-  "D",
-  "L",
-  "U",
-  "M",
-  "W",
-  "C",
-  "F",
-  "G",
-  "Y",
-  "P",
-  "B",
-  "V",
-  "K",
-  "'",
-  "X",
-  "J",
-  "Q",
-  "Z"
-];
+var ARKIT_BLENDSHAPES = LAM_BLENDSHAPES;
 var ARKIT_SYMMETRIC_PAIRS = [
   ["jawLeft", "jawRight"],
   ["mouthLeft", "mouthRight"],
@@ -29240,6 +29209,107 @@ function symmetrizeBlendshapes(frame) {
   }
   return result;
 }
+var WAV2ARKIT_BLENDSHAPES = [
+  "browDownLeft",
+  "browDownRight",
+  "browInnerUp",
+  "browOuterUpLeft",
+  "browOuterUpRight",
+  "cheekPuff",
+  "cheekSquintLeft",
+  "cheekSquintRight",
+  "eyeBlinkLeft",
+  "eyeBlinkRight",
+  "eyeLookDownLeft",
+  "eyeLookDownRight",
+  "eyeLookInLeft",
+  "eyeLookInRight",
+  "eyeLookOutLeft",
+  "eyeLookOutRight",
+  "eyeLookUpLeft",
+  "eyeLookUpRight",
+  "eyeSquintLeft",
+  "eyeSquintRight",
+  "eyeWideLeft",
+  "eyeWideRight",
+  "jawForward",
+  "jawLeft",
+  "jawOpen",
+  "mouthFrownLeft",
+  "mouthFrownRight",
+  "mouthFunnel",
+  "mouthLeft",
+  "mouthLowerDownLeft",
+  "mouthLowerDownRight",
+  "mouthPressLeft",
+  "mouthPressRight",
+  "mouthPucker",
+  "mouthRight",
+  "mouthRollLower",
+  "mouthRollUpper",
+  "mouthShrugLower",
+  "mouthShrugUpper",
+  "mouthSmileLeft",
+  "mouthSmileRight",
+  "mouthStretchLeft",
+  "mouthStretchRight",
+  "mouthUpperUpLeft",
+  "mouthUpperUpRight",
+  "noseSneerLeft",
+  "noseSneerRight",
+  "tongueOut",
+  "mouthClose",
+  "mouthDimpleLeft",
+  "mouthDimpleRight",
+  "jawRight"
+];
+var REMAP_WAV2ARKIT_TO_LAM = WAV2ARKIT_BLENDSHAPES.map(
+  (name) => LAM_BLENDSHAPES.indexOf(name)
+);
+function remapWav2ArkitToLam(frame) {
+  const result = new Float32Array(52);
+  for (let i = 0; i < 52; i++) {
+    result[REMAP_WAV2ARKIT_TO_LAM[i]] = frame[i];
+  }
+  return result;
+}
+
+// src/inference/Wav2Vec2Inference.ts
+var logger2 = createLogger("Wav2Vec2");
+var CTC_VOCAB = [
+  "<pad>",
+  "<s>",
+  "</s>",
+  "<unk>",
+  "|",
+  "E",
+  "T",
+  "A",
+  "O",
+  "N",
+  "I",
+  "H",
+  "S",
+  "R",
+  "D",
+  "L",
+  "U",
+  "M",
+  "W",
+  "C",
+  "F",
+  "G",
+  "Y",
+  "P",
+  "B",
+  "V",
+  "K",
+  "'",
+  "X",
+  "J",
+  "Q",
+  "Z"
+];
 var Wav2Vec2Inference = class {
   constructor(config) {
     this.session = null;
@@ -29276,8 +29346,8 @@ var Wav2Vec2Inference = class {
     });
     try {
       logger2.info("Loading ONNX Runtime...", { preference: this.config.backend || "auto" });
-      const { ort: ort2, backend } = await getOnnxRuntimeForPreference(this.config.backend || "auto");
-      this.ort = ort2;
+      const { ort, backend } = await getOnnxRuntimeForPreference(this.config.backend || "auto");
+      this.ort = ort;
       this._backend = backend;
       logger2.info("ONNX Runtime loaded", { backend: this._backend });
       const cache = getModelCache();
@@ -29478,6 +29548,7 @@ var Wav2Vec2Inference = class {
             blendshapes,
             asrLogits,
             text,
+            numFrames: numA2EFrames,
             numA2EFrames,
             numASRFrames,
             inferenceTimeMs
@@ -29845,8 +29916,293 @@ var WhisperInference = class _WhisperInference {
   }
 };
 
+// src/inference/Wav2ArkitCpuInference.ts
+var logger5 = createLogger("Wav2ArkitCpu");
+var Wav2ArkitCpuInference = class {
+  constructor(config) {
+    this.session = null;
+    this.ort = null;
+    this._backend = "wasm";
+    this.isLoading = false;
+    // Inference queue for handling concurrent calls
+    this.inferenceQueue = Promise.resolve();
+    this.config = config;
+  }
+  get backend() {
+    return this.session ? this._backend : null;
+  }
+  get isLoaded() {
+    return this.session !== null;
+  }
+  /**
+   * Load the ONNX model
+   */
+  async load() {
+    if (this.isLoading) {
+      throw new Error("Model is already loading");
+    }
+    if (this.session) {
+      throw new Error("Model already loaded. Call dispose() first.");
+    }
+    this.isLoading = true;
+    const startTime = performance.now();
+    const telemetry = getTelemetry();
+    const span = telemetry?.startSpan("Wav2ArkitCpu.load", {
+      "model.url": this.config.modelUrl,
+      "model.backend_requested": this.config.backend || "wasm"
+    });
+    try {
+      const preference = this.config.backend || "wasm";
+      logger5.info("Loading ONNX Runtime...", { preference });
+      const { ort, backend } = await getOnnxRuntimeForPreference(preference);
+      this.ort = ort;
+      this._backend = backend;
+      logger5.info("ONNX Runtime loaded", { backend: this._backend });
+      const cache = getModelCache();
+      const modelUrl = this.config.modelUrl;
+      const isCached = await cache.has(modelUrl);
+      let modelBuffer;
+      if (isCached) {
+        logger5.debug("Loading model from cache", { modelUrl });
+        modelBuffer = await cache.get(modelUrl);
+        if (!modelBuffer) {
+          logger5.warn("Cache corruption detected, clearing and retrying", { modelUrl });
+          await cache.delete(modelUrl);
+          modelBuffer = await fetchWithCache(modelUrl);
+        }
+      } else {
+        logger5.debug("Fetching and caching model", { modelUrl });
+        modelBuffer = await fetchWithCache(modelUrl);
+      }
+      if (!modelBuffer) {
+        throw new Error(`Failed to load model: ${modelUrl}`);
+      }
+      logger5.debug("Creating ONNX session", {
+        size: formatBytes(modelBuffer.byteLength),
+        backend: this._backend
+      });
+      const sessionOptions = getSessionOptions(this._backend);
+      const modelData = new Uint8Array(modelBuffer);
+      this.session = await this.ort.InferenceSession.create(modelData, sessionOptions);
+      const loadTimeMs = performance.now() - startTime;
+      logger5.info("Model loaded successfully", {
+        backend: this._backend,
+        loadTimeMs: Math.round(loadTimeMs),
+        inputs: this.session.inputNames,
+        outputs: this.session.outputNames
+      });
+      span?.setAttributes({
+        "model.backend": this._backend,
+        "model.load_time_ms": loadTimeMs,
+        "model.cached": isCached
+      });
+      span?.end();
+      telemetry?.recordHistogram("omote.model.load_time", loadTimeMs, {
+        model: "wav2arkit_cpu",
+        backend: this._backend
+      });
+      logger5.debug("Running warmup inference");
+      const warmupStart = performance.now();
+      const silentAudio = new Float32Array(16e3);
+      await this.infer(silentAudio);
+      const warmupTimeMs = performance.now() - warmupStart;
+      logger5.info("Warmup inference complete", {
+        warmupTimeMs: Math.round(warmupTimeMs),
+        backend: this._backend
+      });
+      telemetry?.recordHistogram("omote.model.warmup_time", warmupTimeMs, {
+        model: "wav2arkit_cpu",
+        backend: this._backend
+      });
+      return {
+        backend: this._backend,
+        loadTimeMs,
+        inputNames: [...this.session.inputNames],
+        outputNames: [...this.session.outputNames]
+      };
+    } catch (error) {
+      span?.endWithError(error instanceof Error ? error : new Error(String(error)));
+      telemetry?.incrementCounter("omote.errors.total", 1, {
+        model: "wav2arkit_cpu",
+        error_type: "load_failed"
+      });
+      throw error;
+    } finally {
+      this.isLoading = false;
+    }
+  }
+  /**
+   * Run inference on raw audio
+   *
+   * Accepts variable-length audio (not fixed to 16000 samples).
+   * Output frames = ceil(30 * numSamples / 16000).
+   *
+   * @param audioSamples - Float32Array of raw audio at 16kHz
+   * @param _identityIndex - Ignored (identity 11 is baked into the model)
+   */
+  async infer(audioSamples, _identityIndex) {
+    if (!this.session) {
+      throw new Error("Model not loaded. Call load() first.");
+    }
+    const audioCopy = new Float32Array(audioSamples);
+    const feeds = {
+      "audio_waveform": new this.ort.Tensor("float32", audioCopy, [1, audioCopy.length])
+    };
+    return this.queueInference(feeds, audioCopy.length);
+  }
+  /**
+   * Queue inference to serialize ONNX session calls
+   */
+  queueInference(feeds, inputSamples) {
+    return new Promise((resolve, reject) => {
+      this.inferenceQueue = this.inferenceQueue.then(async () => {
+        const telemetry = getTelemetry();
+        const span = telemetry?.startSpan("Wav2ArkitCpu.infer", {
+          "inference.backend": this._backend,
+          "inference.input_samples": inputSamples
+        });
+        try {
+          const startTime = performance.now();
+          const results = await this.session.run(feeds);
+          const inferenceTimeMs = performance.now() - startTime;
+          const blendshapeOutput = results["blendshapes"];
+          if (!blendshapeOutput) {
+            throw new Error("Missing blendshapes output from model");
+          }
+          const blendshapeData = blendshapeOutput.data;
+          const numFrames = blendshapeOutput.dims[1];
+          const numBlendshapes = blendshapeOutput.dims[2];
+          const blendshapes = [];
+          for (let f = 0; f < numFrames; f++) {
+            const rawFrame = blendshapeData.slice(f * numBlendshapes, (f + 1) * numBlendshapes);
+            const remapped = remapWav2ArkitToLam(rawFrame);
+            blendshapes.push(symmetrizeBlendshapes(remapped));
+          }
+          logger5.trace("Inference completed", {
+            inferenceTimeMs: Math.round(inferenceTimeMs * 100) / 100,
+            numFrames,
+            inputSamples
+          });
+          span?.setAttributes({
+            "inference.duration_ms": inferenceTimeMs,
+            "inference.frames": numFrames
+          });
+          span?.end();
+          telemetry?.recordHistogram("omote.inference.latency", inferenceTimeMs, {
+            model: "wav2arkit_cpu",
+            backend: this._backend
+          });
+          telemetry?.incrementCounter("omote.inference.total", 1, {
+            model: "wav2arkit_cpu",
+            backend: this._backend,
+            status: "success"
+          });
+          resolve({
+            blendshapes,
+            numFrames,
+            inferenceTimeMs
+          });
+        } catch (err) {
+          span?.endWithError(err instanceof Error ? err : new Error(String(err)));
+          telemetry?.incrementCounter("omote.inference.total", 1, {
+            model: "wav2arkit_cpu",
+            backend: this._backend,
+            status: "error"
+          });
+          reject(err);
+        }
+      });
+    });
+  }
+  /**
+   * Dispose of the model and free resources
+   */
+  async dispose() {
+    if (this.session) {
+      await this.session.release();
+      this.session = null;
+    }
+  }
+};
+
+// src/inference/createLipSync.ts
+var logger6 = createLogger("createLipSync");
+function createLipSync(config) {
+  const mode = config.mode ?? "auto";
+  const fallbackOnError = config.fallbackOnError ?? true;
+  let useCpu;
+  if (mode === "cpu") {
+    useCpu = true;
+    logger6.info("Forcing CPU lip sync model (wav2arkit_cpu)");
+  } else if (mode === "gpu") {
+    useCpu = false;
+    logger6.info("Forcing GPU lip sync model (Wav2Vec2)");
+  } else {
+    useCpu = isSafari();
+    logger6.info("Auto-detected lip sync model", {
+      useCpu,
+      isSafari: isSafari()
+    });
+  }
+  if (useCpu) {
+    logger6.info("Creating Wav2ArkitCpuInference (1.8MB, WASM)");
+    return new Wav2ArkitCpuInference({
+      modelUrl: config.cpuModelUrl
+    });
+  }
+  const gpuInstance = new Wav2Vec2Inference({
+    modelUrl: config.gpuModelUrl,
+    backend: config.gpuBackend ?? "auto",
+    numIdentityClasses: config.numIdentityClasses
+  });
+  if (fallbackOnError) {
+    logger6.info("Creating Wav2Vec2Inference with CPU fallback");
+    return new LipSyncWithFallback(gpuInstance, config);
+  }
+  logger6.info("Creating Wav2Vec2Inference (no fallback)");
+  return gpuInstance;
+}
+var LipSyncWithFallback = class {
+  constructor(gpuInstance, config) {
+    this.hasFallenBack = false;
+    this.implementation = gpuInstance;
+    this.config = config;
+  }
+  get backend() {
+    return this.implementation.backend;
+  }
+  get isLoaded() {
+    return this.implementation.isLoaded;
+  }
+  async load() {
+    try {
+      return await this.implementation.load();
+    } catch (error) {
+      logger6.warn("GPU model load failed, falling back to CPU model", {
+        error: error instanceof Error ? error.message : String(error)
+      });
+      try {
+        await this.implementation.dispose();
+      } catch {
+      }
+      this.implementation = new Wav2ArkitCpuInference({
+        modelUrl: this.config.cpuModelUrl
+      });
+      this.hasFallenBack = true;
+      logger6.info("Fallback to Wav2ArkitCpuInference successful");
+      return await this.implementation.load();
+    }
+  }
+  async infer(audioSamples, identityIndex) {
+    return this.implementation.infer(audioSamples, identityIndex);
+  }
+  async dispose() {
+    return this.implementation.dispose();
+  }
+};
+
 // src/inference/SileroVADInference.ts
-var logger5 = createLogger("SileroVAD");
+var logger7 = createLogger("SileroVAD");
 var SileroVADInference = class {
   constructor(config) {
     this.session = null;
@@ -29918,32 +30274,32 @@ var SileroVADInference = class {
       "model.sample_rate": this.config.sampleRate
     });
     try {
-      logger5.info("Loading ONNX Runtime...", { preference: this.config.backend });
-      const { ort: ort2, backend } = await getOnnxRuntimeForPreference(this.config.backend);
-      this.ort = ort2;
+      logger7.info("Loading ONNX Runtime...", { preference: this.config.backend });
+      const { ort, backend } = await getOnnxRuntimeForPreference(this.config.backend);
+      this.ort = ort;
       this._backend = backend;
-      logger5.info("ONNX Runtime loaded", { backend: this._backend });
+      logger7.info("ONNX Runtime loaded", { backend: this._backend });
       const cache = getModelCache();
       const modelUrl = this.config.modelUrl;
       const isCached = await cache.has(modelUrl);
       let modelBuffer;
       if (isCached) {
-        logger5.debug("Loading model from cache", { modelUrl });
+        logger7.debug("Loading model from cache", { modelUrl });
         modelBuffer = await cache.get(modelUrl);
       } else {
-        logger5.debug("Fetching and caching model", { modelUrl });
+        logger7.debug("Fetching and caching model", { modelUrl });
         modelBuffer = await fetchWithCache(modelUrl);
       }
-      logger5.debug("Creating ONNX session", {
+      logger7.debug("Creating ONNX session", {
         size: formatBytes(modelBuffer.byteLength),
         backend: this._backend
       });
       const sessionOptions = getSessionOptions(this._backend);
       const modelData = new Uint8Array(modelBuffer);
-      this.session = await ort2.InferenceSession.create(modelData, sessionOptions);
+      this.session = await ort.InferenceSession.create(modelData, sessionOptions);
       this.reset();
       const loadTimeMs = performance.now() - startTime;
-      logger5.info("Model loaded successfully", {
+      logger7.info("Model loaded successfully", {
         backend: this._backend,
         loadTimeMs: Math.round(loadTimeMs),
         sampleRate: this.config.sampleRate,
@@ -30096,7 +30452,7 @@ var SileroVADInference = class {
           this.preSpeechBuffer.shift();
         }
       }
-      logger5.trace("Skipping VAD inference - audio too quiet", {
+      logger7.trace("Skipping VAD inference - audio too quiet", {
         rms: Math.round(rms * 1e4) / 1e4,
         threshold: MIN_ENERGY_THRESHOLD
       });
@@ -30150,7 +30506,7 @@ var SileroVADInference = class {
           if (isSpeech && !this.wasSpeaking) {
             preSpeechChunks = [...this.preSpeechBuffer];
             this.preSpeechBuffer = [];
-            logger5.debug("Speech started with pre-speech buffer", {
+            logger7.debug("Speech started with pre-speech buffer", {
               preSpeechChunks: preSpeechChunks.length,
               durationMs: Math.round(preSpeechChunks.length * this.getChunkDurationMs())
             });
@@ -30163,7 +30519,7 @@ var SileroVADInference = class {
             this.preSpeechBuffer = [];
           }
           this.wasSpeaking = isSpeech;
-          logger5.trace("VAD inference completed", {
+          logger7.trace("VAD inference completed", {
             probability: Math.round(probability * 1e3) / 1e3,
             isSpeech,
             inferenceTimeMs: Math.round(inferenceTimeMs * 100) / 100
@@ -30219,7 +30575,7 @@ var SileroVADInference = class {
 SileroVADInference.isWebGPUAvailable = isWebGPUAvailable;
 
 // src/inference/SileroVADWorker.ts
-var logger6 = createLogger("SileroVADWorker");
+var logger8 = createLogger("SileroVADWorker");
 var WASM_CDN_PATH2 = "https://cdn.jsdelivr.net/npm/onnxruntime-web@1.23.2/dist/";
 var LOAD_TIMEOUT_MS = 1e4;
 var INFERENCE_TIMEOUT_MS = 1e3;
@@ -30482,7 +30838,7 @@ var SileroVADWorker = class {
       this.handleWorkerMessage(event.data);
     };
     worker.onerror = (error) => {
-      logger6.error("Worker error", { error: error.message });
+      logger8.error("Worker error", { error: error.message });
       for (const [, resolver] of this.pendingResolvers) {
         resolver.reject(new Error(`Worker error: ${error.message}`));
       }
@@ -30558,9 +30914,9 @@ var SileroVADWorker = class {
       "model.sample_rate": this.config.sampleRate
     });
     try {
-      logger6.info("Creating VAD worker...");
+      logger8.info("Creating VAD worker...");
       this.worker = this.createWorker();
-      logger6.info("Loading model in worker...", {
+      logger8.info("Loading model in worker...", {
         modelUrl: this.config.modelUrl,
         sampleRate: this.config.sampleRate
       });
@@ -30576,7 +30932,7 @@ var SileroVADWorker = class {
       );
       this._isLoaded = true;
       const loadTimeMs = performance.now() - startTime;
-      logger6.info("VAD worker loaded successfully", {
+      logger8.info("VAD worker loaded successfully", {
         backend: "wasm",
         loadTimeMs: Math.round(loadTimeMs),
         workerLoadTimeMs: Math.round(result.loadTimeMs),
@@ -30683,7 +31039,7 @@ var SileroVADWorker = class {
           if (isSpeech && !this.wasSpeaking) {
             preSpeechChunks = [...this.preSpeechBuffer];
             this.preSpeechBuffer = [];
-            logger6.debug("Speech started with pre-speech buffer", {
+            logger8.debug("Speech started with pre-speech buffer", {
               preSpeechChunks: preSpeechChunks.length,
               durationMs: Math.round(preSpeechChunks.length * this.getChunkDurationMs())
             });
@@ -30696,7 +31052,7 @@ var SileroVADWorker = class {
             this.preSpeechBuffer = [];
           }
           this.wasSpeaking = isSpeech;
-          logger6.trace("VAD worker inference completed", {
+          logger8.trace("VAD worker inference completed", {
             probability: Math.round(result.probability * 1e3) / 1e3,
             isSpeech,
             inferenceTimeMs: Math.round(inferenceTimeMs * 100) / 100,
@@ -30764,18 +31120,18 @@ var SileroVADWorker = class {
 };
 
 // src/inference/createSileroVAD.ts
-var logger7 = createLogger("createSileroVAD");
+var logger9 = createLogger("createSileroVAD");
 function supportsVADWorker() {
   if (typeof Worker === "undefined") {
-    logger7.debug("Worker not supported: Worker constructor undefined");
+    logger9.debug("Worker not supported: Worker constructor undefined");
     return false;
   }
   if (typeof URL === "undefined" || typeof URL.createObjectURL === "undefined") {
-    logger7.debug("Worker not supported: URL.createObjectURL unavailable");
+    logger9.debug("Worker not supported: URL.createObjectURL unavailable");
     return false;
   }
   if (typeof Blob === "undefined") {
-    logger7.debug("Worker not supported: Blob constructor unavailable");
+    logger9.debug("Worker not supported: Blob constructor unavailable");
     return false;
   }
   return true;
@@ -30785,19 +31141,19 @@ function createSileroVAD(config) {
   let useWorker;
   if (config.useWorker !== void 0) {
     useWorker = config.useWorker;
-    logger7.debug("Worker preference explicitly set", { useWorker });
+    logger9.debug("Worker preference explicitly set", { useWorker });
   } else {
     const workerSupported = supportsVADWorker();
     const onMobile = isMobile();
     useWorker = workerSupported && !onMobile;
-    logger7.debug("Auto-detected Worker preference", {
+    logger9.debug("Auto-detected Worker preference", {
       useWorker,
       workerSupported,
       onMobile
     });
   }
   if (useWorker) {
-    logger7.info("Creating SileroVADWorker (off-main-thread)");
+    logger9.info("Creating SileroVADWorker (off-main-thread)");
     const worker = new SileroVADWorker({
       modelUrl: config.modelUrl,
       sampleRate: config.sampleRate,
@@ -30809,7 +31165,7 @@ function createSileroVAD(config) {
     }
     return worker;
   }
-  logger7.info("Creating SileroVADInference (main thread)");
+  logger9.info("Creating SileroVADInference (main thread)");
   return new SileroVADInference(config);
 }
 var VADWorkerWithFallback = class {
@@ -30835,7 +31191,7 @@ var VADWorkerWithFallback = class {
     try {
       return await this.implementation.load();
     } catch (error) {
-      logger7.warn("Worker load failed, falling back to main thread", {
+      logger9.warn("Worker load failed, falling back to main thread", {
         error: error instanceof Error ? error.message : String(error)
       });
       try {
@@ -30844,7 +31200,7 @@ var VADWorkerWithFallback = class {
       }
       this.implementation = new SileroVADInference(this.config);
       this.hasFallenBack = true;
-      logger7.info("Fallback to SileroVADInference successful");
+      logger9.info("Fallback to SileroVADInference successful");
       return await this.implementation.load();
     }
   }
@@ -30866,7 +31222,7 @@ var VADWorkerWithFallback = class {
 };
 
 // src/inference/Emotion2VecInference.ts
-var logger8 = createLogger("Emotion2Vec");
+var logger10 = createLogger("Emotion2Vec");
 var EMOTION2VEC_LABELS = ["neutral", "happy", "angry", "sad"];
 var Emotion2VecInference = class {
   constructor(config) {
@@ -30908,36 +31264,36 @@ var Emotion2VecInference = class {
       "model.backend_requested": this.config.backend
     });
     try {
-      logger8.info("Loading ONNX Runtime...", { preference: this.config.backend });
-      const { ort: ort2, backend } = await getOnnxRuntimeForPreference(this.config.backend);
-      this.ort = ort2;
+      logger10.info("Loading ONNX Runtime...", { preference: this.config.backend });
+      const { ort, backend } = await getOnnxRuntimeForPreference(this.config.backend);
+      this.ort = ort;
       this._backend = backend;
-      logger8.info("ONNX Runtime loaded", { backend: this._backend });
-      logger8.info("Checking model cache...");
+      logger10.info("ONNX Runtime loaded", { backend: this._backend });
+      logger10.info("Checking model cache...");
       const cache = getModelCache();
       const modelUrl = this.config.modelUrl;
       const isCached = await cache.has(modelUrl);
-      logger8.info("Cache check complete", { modelUrl, isCached });
+      logger10.info("Cache check complete", { modelUrl, isCached });
       let modelBuffer;
       if (isCached) {
-        logger8.info("Loading model from cache...", { modelUrl });
+        logger10.info("Loading model from cache...", { modelUrl });
         modelBuffer = await cache.get(modelUrl);
-        logger8.info("Model loaded from cache", { size: formatBytes(modelBuffer.byteLength) });
+        logger10.info("Model loaded from cache", { size: formatBytes(modelBuffer.byteLength) });
       } else {
-        logger8.info("Fetching model (not cached)...", { modelUrl });
+        logger10.info("Fetching model (not cached)...", { modelUrl });
         modelBuffer = await fetchWithCache(modelUrl);
-        logger8.info("Model fetched and cached", { size: formatBytes(modelBuffer.byteLength) });
+        logger10.info("Model fetched and cached", { size: formatBytes(modelBuffer.byteLength) });
       }
-      logger8.info("Creating ONNX session (this may take a while for large models)...");
-      logger8.debug("Creating ONNX session", {
+      logger10.info("Creating ONNX session (this may take a while for large models)...");
+      logger10.debug("Creating ONNX session", {
         size: formatBytes(modelBuffer.byteLength),
         backend: this._backend
       });
       const sessionOptions = getSessionOptions(this._backend);
       const modelData = new Uint8Array(modelBuffer);
-      this.session = await ort2.InferenceSession.create(modelData, sessionOptions);
+      this.session = await ort.InferenceSession.create(modelData, sessionOptions);
       const loadTimeMs = performance.now() - startTime;
-      logger8.info("Model loaded successfully", {
+      logger10.info("Model loaded successfully", {
         backend: this._backend,
         loadTimeMs: Math.round(loadTimeMs),
         sampleRate: this.config.sampleRate,
@@ -31049,7 +31405,7 @@ var Emotion2VecInference = class {
             });
           }
           const inferenceTimeMs = performance.now() - startTime;
-          logger8.debug("Emotion inference completed", {
+          logger10.debug("Emotion inference completed", {
             numFrames,
             dominant: dominant.emotion,
             confidence: Math.round(dominant.confidence * 100),
@@ -31125,15 +31481,8 @@ var Emotion2VecInference = class {
  */
 Emotion2VecInference.isWebGPUAvailable = isWebGPUAvailable;
 
-// src/inference/ChatterboxTurboInference.ts
-var import_webgpu = __toESM(require("onnxruntime-web/webgpu"));
-init_transformers_web();
-var logger9 = createLogger("ChatterboxTurbo");
-env2.allowLocalModels = true;
-import_webgpu.default.env.wasm.wasmPaths = "https://cdn.jsdelivr.net/npm/onnxruntime-web@1.23.2/dist/";
-
 // src/inference/SafariSpeechRecognition.ts
-var logger10 = createLogger("SafariSpeech");
+var logger11 = createLogger("SafariSpeech");
 var SafariSpeechRecognition = class _SafariSpeechRecognition {
   constructor(config = {}) {
     this.recognition = null;
@@ -31152,7 +31501,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
       interimResults: config.interimResults ?? true,
       maxAlternatives: config.maxAlternatives ?? 1
     };
-    logger10.debug("SafariSpeechRecognition created", {
+    logger11.debug("SafariSpeechRecognition created", {
       language: this.config.language,
       continuous: this.config.continuous
     });
@@ -31213,7 +31562,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
    */
   async start() {
     if (this.isListening) {
-      logger10.warn("Already listening");
+      logger11.warn("Already listening");
       return;
     }
     if (!_SafariSpeechRecognition.isAvailable()) {
@@ -31243,7 +31592,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
       this.isListening = true;
       this.startTime = performance.now();
       this.accumulatedText = "";
-      logger10.info("Speech recognition started", {
+      logger11.info("Speech recognition started", {
         language: this.config.language
       });
       span?.end();
@@ -31258,7 +31607,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
    */
   async stop() {
     if (!this.isListening || !this.recognition) {
-      logger10.warn("Not currently listening");
+      logger11.warn("Not currently listening");
       return {
         text: this.accumulatedText,
         language: this.config.language,
@@ -31287,7 +31636,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
     if (this.recognition && this.isListening) {
       this.recognition.abort();
       this.isListening = false;
-      logger10.info("Speech recognition aborted");
+      logger11.info("Speech recognition aborted");
     }
   }
   /**
@@ -31318,7 +31667,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
     this.isListening = false;
     this.resultCallbacks = [];
     this.errorCallbacks = [];
-    logger10.debug("SafariSpeechRecognition disposed");
+    logger11.debug("SafariSpeechRecognition disposed");
   }
   /**
    * Set up event handlers for the recognition instance
@@ -31346,7 +31695,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
               confidence: alternative.confidence
             };
             this.emitResult(speechResult);
-            logger10.trace("Speech result", {
+            logger11.trace("Speech result", {
               text: text.substring(0, 50),
               isFinal,
               confidence: alternative.confidence
@@ -31356,12 +31705,12 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
         span?.end();
       } catch (error) {
         span?.endWithError(error instanceof Error ? error : new Error(String(error)));
-        logger10.error("Error processing speech result", { error });
+        logger11.error("Error processing speech result", { error });
       }
     };
     this.recognition.onerror = (event) => {
       const error = new Error(`Speech recognition error: ${event.error} - ${event.message}`);
-      logger10.error("Speech recognition error", { error: event.error, message: event.message });
+      logger11.error("Speech recognition error", { error: event.error, message: event.message });
       this.emitError(error);
       if (this.stopRejecter) {
         this.stopRejecter(error);
@@ -31371,7 +31720,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
     };
     this.recognition.onend = () => {
       this.isListening = false;
-      logger10.info("Speech recognition ended", {
+      logger11.info("Speech recognition ended", {
         totalText: this.accumulatedText.length,
         durationMs: performance.now() - this.startTime
       });
@@ -31388,13 +31737,13 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
       }
     };
     this.recognition.onstart = () => {
-      logger10.debug("Speech recognition started by browser");
+      logger11.debug("Speech recognition started by browser");
     };
     this.recognition.onspeechstart = () => {
-      logger10.debug("Speech detected");
+      logger11.debug("Speech detected");
     };
     this.recognition.onspeechend = () => {
-      logger10.debug("Speech ended");
+      logger11.debug("Speech ended");
     };
   }
   /**
@@ -31405,7 +31754,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
       try {
         callback(result);
       } catch (error) {
-        logger10.error("Error in result callback", { error });
+        logger11.error("Error in result callback", { error });
       }
     }
   }
@@ -31417,7 +31766,7 @@ var SafariSpeechRecognition = class _SafariSpeechRecognition {
       try {
         callback(error);
       } catch (callbackError) {
-        logger10.error("Error in error callback", { error: callbackError });
+        logger11.error("Error in error callback", { error: callbackError });
       }
     }
   }
@@ -32840,12 +33189,12 @@ async function isHuggingFaceCDNReachable(testUrl = HF_CDN_TEST_URL) {
 }
 
 // src/utils/transformersCacheClear.ts
-var logger11 = createLogger("TransformersCache");
+var logger12 = createLogger("TransformersCache");
 async function clearTransformersCache(options) {
   const verbose = options?.verbose ?? true;
   const additionalPatterns = options?.additionalPatterns ?? [];
   if (!("caches" in window)) {
-    logger11.warn("Cache API not available in this environment");
+    logger12.warn("Cache API not available in this environment");
     return [];
   }
   try {
@@ -32863,18 +33212,18 @@ async function clearTransformersCache(options) {
       );
       if (shouldDelete) {
         if (verbose) {
-          logger11.info("Deleting cache", { cacheName });
+          logger12.info("Deleting cache", { cacheName });
         }
         const deleted = await caches.delete(cacheName);
         if (deleted) {
           deletedCaches.push(cacheName);
         } else if (verbose) {
-          logger11.warn("Failed to delete cache", { cacheName });
+          logger12.warn("Failed to delete cache", { cacheName });
         }
       }
     }
     if (verbose) {
-      logger11.info("Cache clearing complete", {
+      logger12.info("Cache clearing complete", {
         totalCaches: cacheNames.length,
         deletedCount: deletedCaches.length,
         deletedCaches
@@ -32882,35 +33231,35 @@ async function clearTransformersCache(options) {
     }
     return deletedCaches;
   } catch (error) {
-    logger11.error("Error clearing caches", { error });
+    logger12.error("Error clearing caches", { error });
     throw error;
   }
 }
 async function clearSpecificCache(cacheName) {
   if (!("caches" in window)) {
-    logger11.warn("Cache API not available in this environment");
+    logger12.warn("Cache API not available in this environment");
     return false;
   }
   try {
     const deleted = await caches.delete(cacheName);
-    logger11.info("Cache deletion attempt", { cacheName, deleted });
+    logger12.info("Cache deletion attempt", { cacheName, deleted });
     return deleted;
   } catch (error) {
-    logger11.error("Error deleting cache", { cacheName, error });
+    logger12.error("Error deleting cache", { cacheName, error });
     return false;
   }
 }
 async function listCaches() {
   if (!("caches" in window)) {
-    logger11.warn("Cache API not available in this environment");
+    logger12.warn("Cache API not available in this environment");
     return [];
   }
   try {
     const cacheNames = await caches.keys();
-    logger11.debug("Available caches", { cacheNames });
+    logger12.debug("Available caches", { cacheNames });
     return cacheNames;
   } catch (error) {
-    logger11.error("Error listing caches", { error });
+    logger12.error("Error listing caches", { error });
     return [];
   }
 }
@@ -32952,7 +33301,7 @@ async function validateCachedResponse(cacheName, requestUrl) {
       reason: valid ? "Valid response" : `Invalid: status=${response.status}, contentType=${contentType}, isHtml=${isHtml || looksLikeHtml}`
     };
   } catch (error) {
-    logger11.error("Error validating cached response", { cacheName, requestUrl, error });
+    logger12.error("Error validating cached response", { cacheName, requestUrl, error });
     return {
       exists: false,
       valid: false,
@@ -32989,7 +33338,7 @@ async function scanForInvalidCaches() {
         }
       }
     }
-    logger11.info("Cache scan complete", {
+    logger12.info("Cache scan complete", {
       totalCaches: cacheNames.length,
       scannedEntries,
       invalidCount: invalidEntries.length
@@ -33000,13 +33349,13 @@ async function scanForInvalidCaches() {
       invalidEntries
     };
   } catch (error) {
-    logger11.error("Error scanning caches", { error });
+    logger12.error("Error scanning caches", { error });
     throw error;
   }
 }
 async function nukeBrowserCaches(preventRecreation = false) {
   if (!("caches" in window)) {
-    logger11.warn("Cache API not available in this environment");
+    logger12.warn("Cache API not available in this environment");
     return 0;
   }
   try {
@@ -33018,17 +33367,17 @@ async function nukeBrowserCaches(preventRecreation = false) {
         deletedCount++;
       }
     }
-    logger11.info("All browser caches cleared", {
+    logger12.info("All browser caches cleared", {
       totalDeleted: deletedCount
     });
     if (preventRecreation) {
       const { env: env3 } = await Promise.resolve().then(() => (init_transformers_web(), transformers_web_exports));
       env3.useBrowserCache = false;
-      logger11.warn("Browser cache creation disabled (env.useBrowserCache = false)");
+      logger12.warn("Browser cache creation disabled (env.useBrowserCache = false)");
     }
     return deletedCount;
   } catch (error) {
-    logger11.error("Error nuking caches", { error });
+    logger12.error("Error nuking caches", { error });
     throw error;
   }
 }