@fugood/bricks-ctor 2.24.0-beta.40

package/types/generators/TextToSpeechGgml.ts

@@ -0,0 +1,221 @@
+/* Auto generated by build script
+ *
+ * Local Text-to-Speech (TTS) inference based on GGML and [llama.cpp](https://github.com/ggerganov/llama.cpp)
+ *  You can use any converted model on HuggingFace.
+ *
+ *  ## Notice
+ *  - The device RAM must be larger than 8GB
+ *  - iOS: Recommended use M1+ / A17+ chip device. Supported GPU acceleration by Metal.
+ *  - macOS: Recommended use M1+ chip device. Supported GPU acceleration by Metal.
+ *  - Android: Recommended use Android 13+ system.
+ *    - Supported GPU acceleration by OpenCL, currently only for Qualcomm Adreno 700+ GPUs, other GPUs are not supported.
+ *    - Supported Hexagon NPU for Qualcomm Snapdragon 8 Gen 2+ GPUs.
+ *  - Linux / Windows
+ *    - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property
+ *    - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux)
+ */
+import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
+import type { Data, DataLink } from '../data'
+import type {
+  Brick,
+  Generator,
+  EventAction,
+  ActionWithDataParams,
+  ActionWithParams,
+  Action,
+  EventProperty,
+} from '../common'
+import type { TemplateEventPropsMap } from '../../utils/event-props'
+
+/* Load the model */
+export type GeneratorGGMLTTSActionLoadModel = Action & {
+  __actionName: 'GENERATOR_GGML_TTS_LOAD_MODEL'
+}
+
+/* Generate audio */
+export type GeneratorGGMLTTSActionGenerate = ActionWithParams & {
+  __actionName: 'GENERATOR_GGML_TTS_GENERATE'
+  params?: Array<{
+    input: 'text'
+    value?: string | DataLink | EventProperty
+    mapping?: string
+  }>
+}
+
+/* Clean cache */
+export type GeneratorGGMLTTSActionCleanCache = Action & {
+  __actionName: 'GENERATOR_GGML_TTS_CLEAN_CACHE'
+}
+
+/* Release context */
+export type GeneratorGGMLTTSActionReleaseContext = Action & {
+  __actionName: 'GENERATOR_GGML_TTS_RELEASE_CONTEXT'
+}
+
+/* Cancel current generation and clear queue */
+export type GeneratorGGMLTTSActionCancel = Action & {
+  __actionName: 'GENERATOR_GGML_TTS_CANCEL'
+}
+
+interface GeneratorGGMLTTSDef {
+  /*
+Default property:
+{
+  "vocoderUrl": "https://huggingface.co/ggml-org/WavTokenizer/resolve/main/WavTokenizer-Large-75-F16.gguf",
+  "vocoderHashType": "sha256",
+  "vocoderHash": "2356baa8631cc2995ea3465196a017a2733600d849a91180c0f97fa7fb375bbe",
+  "vocoderBatchSize": 4096,
+  "outputType": "play",
+  "cacheGenerated": true,
+  "autoInferEnable": false,
+  "softBreakRegex": "^[^\\r\\n\\t\\f\\v]*([\\r\\n]+|[。！？!?.]\\B)",
+  "hardBreakTime": 500,
+  "completionTemperature": 0.1,
+  "completionRepetitionPenalty": 1.1,
+  "completionTopK": 40,
+  "completionTopP": 0.9,
+  "completionMinP": 0.05,
+  "useGuideToken": true,
+  "contextSize": 8192,
+  "batchSize": 8192,
+  "microBatchSize": 512,
+  "maxThreads": 2,
+  "accelVariant": "default",
+  "mainGpu": 0,
+  "gpuLayers": 0,
+  "useMlock": true,
+  "useMmap": true,
+  "useFlashAttn": "off"
+}
+  */
+  property?: {
+    /* Initialize the TTS context on generator initialization */
+    init?: boolean | DataLink
+    /* The URL or path of model
+ We used GGUF format model, please refer to https://github.com/ggerganov/llama.cpp/tree/master#description */
+    modelUrl?: string | DataLink
+    /* Hash type of model */
+    modelHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Hash of model */
+    modelHash?: string | DataLink
+    /* The URL or path of vocoder model */
+    vocoderUrl?: string | DataLink
+    /* Hash type of vocoder model */
+    vocoderHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Hash of vocoder model */
+    vocoderHash?: string | DataLink
+    /* Batch size of vocoder model */
+    vocoderBatchSize?: number | DataLink
+    /* Output mode */
+    outputType?: 'play' | 'file' | DataLink
+    /* Enable cache for generated audio */
+    cacheGenerated?: boolean | DataLink
+    /* Text to generate */
+    prompt?: string | DataLink
+    /* Speaker JSON */
+    speaker?: {} | DataLink
+    /* Auto inference when prompt changes */
+    autoInferEnable?: boolean | DataLink
+    /* Segmentation rule for auto inference */
+    softBreakRegex?: string | DataLink
+    /* Time to force inference when softBreakRegex is not satisfied */
+    hardBreakTime?: number | DataLink
+    /* Temperature */
+    completionTemperature?: number | DataLink
+    /* Repetition Penalty */
+    completionRepetitionPenalty?: number | DataLink
+    /* Top K sampling */
+    completionTopK?: number | DataLink
+    /* Top P sampling */
+    completionTopP?: number | DataLink
+    /* Min P sampling */
+    completionMinP?: number | DataLink
+    /* Set the random number generator (RNG) seed (default: -1, -1 = random seed) */
+    completionSeed?: number | DataLink
+    /* Number of tokens to predict */
+    completionPredict?: number | DataLink
+    /* Enable guide token to help prevent hallucinations by forcing the TTS to use the correct words. */
+    useGuideToken?: boolean | DataLink
+    /* Context size, for OutTTS recommended 4096 ~ 8192 (Default to 4096) */
+    contextSize?: number | DataLink
+    /* Logical batch size for prompt processing */
+    batchSize?: number | DataLink
+    /* Physical batch size for prompt processing */
+    microBatchSize?: number | DataLink
+    /* Number of threads */
+    maxThreads?: number | DataLink
+    /* Accelerator variant (Only for desktop)
+ `default` - CPU / Metal (macOS)  
+ `vulkan` - Use Vulkan  
+ `cuda` - Use CUDA  
+ `snapdragon` - Use OpenCL/Hexagon of Qualcomm Snapdragon */
+    accelVariant?: 'default' | 'vulkan' | 'cuda' | 'snapdragon' | DataLink
+    /* Devices. For example:
+  
+ Metal or CPU for iOS/tvOS/MacOS  
+ OpenCL or CPU for Android  
+   - Add `HTP0`, `HTP1`, `...` for OpenCL/Hexagon devices (Use HTP* for all HTP devices)  
+ For Desktop, you may want to include/exclude GPUs */
+    devices?: Array<string | DataLink> | DataLink
+    /* Main GPU index */
+    mainGpu?: number | DataLink
+    /* Number of GPU layers (NOTE: Currently not supported for Android) */
+    gpuLayers?: number | DataLink
+    /* Use memory lock */
+    useMlock?: boolean | DataLink
+    /* Use mmap */
+    useMmap?: boolean | DataLink
+    /* Use Flash Attention for inference (Recommended with GPU enabled) */
+    useFlashAttn?: 'auto' | 'on' | 'off' | DataLink
+  }
+  events?: {
+    /* Event triggered when state change */
+    onContextStateChange?: Array<
+      EventAction<string & keyof TemplateEventPropsMap['GgmlTts']['onContextStateChange']>
+    >
+    /* Event triggered when error occurs */
+    onError?: Array<EventAction<string & keyof TemplateEventPropsMap['GgmlTts']['onError']>>
+  }
+  outlets?: {
+    /* Context state */
+    contextState?: () => Data<string>
+    /* Generated audio file */
+    generatedAudio?: () => Data<string>
+    /* Generated audio file is playing */
+    generatedAudioPlaying?: () => Data<boolean>
+  }
+}
+
+/* Local Text-to-Speech (TTS) inference based on GGML and [llama.cpp](https://github.com/ggerganov/llama.cpp)
+ You can use any converted model on HuggingFace.  
+
+ ## Notice  
+ - The device RAM must be larger than 8GB  
+ - iOS: Recommended use M1+ / A17+ chip device. Supported GPU acceleration by Metal.  
+ - macOS: Recommended use M1+ chip device. Supported GPU acceleration by Metal.  
+ - Android: Recommended use Android 13+ system.  
+   - Supported GPU acceleration by OpenCL, currently only for Qualcomm Adreno 700+ GPUs, other GPUs are not supported.  
+   - Supported Hexagon NPU for Qualcomm Snapdragon 8 Gen 2+ GPUs.  
+ - Linux / Windows  
+   - Supported GPU acceleration, you can choose `vulkan` or `cuda` backend in Accel Variant property  
+   - Supported Hexagon NPU for Qualcomm Dragonwing IQ9 series+ (Linux) */
+export type GeneratorGGMLTTS = Generator &
+  GeneratorGGMLTTSDef & {
+    templateKey: 'GENERATOR_GGML_TTS'
+    switches?: Array<
+      SwitchDef &
+        GeneratorGGMLTTSDef & {
+          conds?: Array<{
+            method: '==' | '!=' | '>' | '<' | '>=' | '<='
+            cond:
+              | SwitchCondInnerStateCurrentCanvas
+              | SwitchCondData
+              | {
+                  __typename: 'SwitchCondInnerStateOutlet'
+                  outlet: 'contextState' | 'generatedAudio' | 'generatedAudioPlaying'
+                  value: any
+                }
+          }>
+        }
+    >
+  }

package/types/generators/TextToSpeechOnnx.ts

@@ -0,0 +1,178 @@
+/* Auto generated by build script
+ *
+ * Local Text-to-Speech (TTS) inference based on ONNX Runtime and [transformers.js](https://huggingface.co/docs/transformers.js)
+ *  You can use any converted model on HuggingFace.
+ */
+import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
+import type { Data, DataLink } from '../data'
+import type {
+  Brick,
+  Generator,
+  EventAction,
+  ActionWithDataParams,
+  ActionWithParams,
+  Action,
+  EventProperty,
+} from '../common'
+import type { TemplateEventPropsMap } from '../../utils/event-props'
+
+/* Load the model */
+export type GeneratorTTSActionLoadModel = Action & {
+  __actionName: 'GENERATOR_TTS_LOAD_MODEL'
+}
+
+/* Generate audio */
+export type GeneratorTTSActionGenerate = ActionWithParams & {
+  __actionName: 'GENERATOR_TTS_GENERATE'
+  params?: Array<{
+    input: 'text'
+    value?: string | DataLink | EventProperty
+    mapping?: string
+  }>
+}
+
+/* Clean cache */
+export type GeneratorTTSActionCleanCache = Action & {
+  __actionName: 'GENERATOR_TTS_CLEAN_CACHE'
+}
+
+/* Release context */
+export type GeneratorTTSActionReleaseContext = Action & {
+  __actionName: 'GENERATOR_TTS_RELEASE_CONTEXT'
+}
+
+/* Cancel current generation and clear queue */
+export type GeneratorTTSActionCancel = Action & {
+  __actionName: 'GENERATOR_TTS_CANCEL'
+}
+
+interface GeneratorTTSDef {
+  /*
+Default property:
+{
+  "model": "BricksDisplay/vits-eng",
+  "modelType": "auto",
+  "vocoderModel": "speecht5_hifigan",
+  "maxLength": 4096,
+  "temperature": 0.1,
+  "repetitionPenalty": 1.1,
+  "doSample": true,
+  "outputType": "play",
+  "cacheGenerated": true,
+  "speed": 1,
+  "autoInferEnable": false,
+  "softBreakRegex": "^[^\\r\\n\\t\\f\\v]*([\\r\\n]+|[。！？!?.]\\B)",
+  "hardBreakTime": 500,
+  "executionMode": "sequential"
+}
+  */
+  property?: {
+    /* Initialize the TTS context on generator initialization */
+    init?: boolean | DataLink
+    /* TTS model
+ The mms-tts models are licensed under CC-BY-NC-4.0 */
+    model?: string | DataLink
+    /* Model type */
+    modelType?: string | DataLink
+    /* Quantize type */
+    quantizeType?:
+      | 'auto'
+      | 'none'
+      | 'fp16'
+      | 'q8'
+      | 'int8'
+      | 'uint8'
+      | 'q4'
+      | 'bnb4'
+      | 'q4f16'
+      | DataLink
+    /* Vocoder model for SpeechT5 */
+    vocoderModel?: 'Custom' | 'speecht5_hifigan' | DataLink
+    /* Custom vocoder model
+ Choose model from https://huggingface.co/models?library=transformers.js&other=hifigan */
+    customVocoderModel?: string | DataLink
+    /* Speaker embedding, for SpeechT5 or StyleTTS (Kokoro) */
+    speakerEmbedUrl?: string | DataLink
+    /* Hash of `speakerEmbedUrl` */
+    speakerEmbedHash?: string | DataLink
+    /* Hash type of `speakerEmbedUrl` */
+    speakerEmbedHashType?: 'md5' | 'sha256' | 'sha1' | DataLink
+    /* Speaker config, for OuteTTS model */
+    speakerConfig?: {} | DataLink
+    /* Audio token generation max length */
+    maxLength?: number | DataLink
+    /* Audio token generation temperature */
+    temperature?: number | DataLink
+    /* Audio token generation repetition penalty */
+    repetitionPenalty?: number | DataLink
+    /* Use greedy sampling for audio token generation */
+    doSample?: boolean | DataLink
+    /* Output mode */
+    outputType?: 'play' | 'file' | DataLink
+    /* Enable cache for generated audio */
+    cacheGenerated?: boolean | DataLink
+    /* Speed of the generated audio, for StyleTTS (Kokoro) */
+    speed?: number | DataLink
+    /* Text to generate */
+    prompt?: string | DataLink
+    /* Auto inference when prompt changes */
+    autoInferEnable?: boolean | DataLink
+    /* Segmentation rule for auto inference */
+    softBreakRegex?: string | DataLink
+    /* Time to force inference when softBreakRegex is not satisfied */
+    hardBreakTime?: number | DataLink
+    /* Executor candidates, descending order of priority
+  Default will be xnnpack, wasm, cpu */
+    executors?:
+      | Array<'qnn' | 'dml' | 'nnapi' | 'xnnpack' | 'coreml' | 'cpu' | 'wasm' | 'webgpu' | DataLink>
+      | DataLink
+    /* Execution mode
+  Usually when the model has many branches, setting this option to `parallel` will give you better performance. */
+    executionMode?: 'sequential' | 'parallel' | DataLink
+    /* QNN backend */
+    qnnBackend?: 'HTP' | 'HTA' | 'DSP' | 'GPU' | 'CPU' | DataLink
+    /* Enable FP16 for QNN HTP */
+    qnnHtpEnableFp16?: boolean | DataLink
+    /* Enable QNN debug */
+    qnnEnableDebug?: boolean | DataLink
+  }
+  events?: {
+    /* Event triggered when state change */
+    onContextStateChange?: Array<
+      EventAction<string & keyof TemplateEventPropsMap['Tts']['onContextStateChange']>
+    >
+    /* Event triggered when error occurs */
+    onError?: Array<EventAction<string & keyof TemplateEventPropsMap['Tts']['onError']>>
+  }
+  outlets?: {
+    /* Context state */
+    contextState?: () => Data<string>
+    /* Generated audio file */
+    generatedAudio?: () => Data<string>
+    /* Generated audio file is playing */
+    generatedAudioPlaying?: () => Data<boolean>
+  }
+}
+
+/* Local Text-to-Speech (TTS) inference based on ONNX Runtime and [transformers.js](https://huggingface.co/docs/transformers.js)
+ You can use any converted model on HuggingFace. */
+export type GeneratorTTS = Generator &
+  GeneratorTTSDef & {
+    templateKey: 'GENERATOR_TTS'
+    switches?: Array<
+      SwitchDef &
+        GeneratorTTSDef & {
+          conds?: Array<{
+            method: '==' | '!=' | '>' | '<' | '>=' | '<='
+            cond:
+              | SwitchCondInnerStateCurrentCanvas
+              | SwitchCondData
+              | {
+                  __typename: 'SwitchCondInnerStateOutlet'
+                  outlet: 'contextState' | 'generatedAudio' | 'generatedAudioPlaying'
+                  value: any
+                }
+          }>
+        }
+    >
+  }

package/types/generators/TextToSpeechOpenAiLike.ts

@@ -0,0 +1,121 @@
+/* Auto generated by build script
+ *
+ * Generate speech from text using OpenAI's Text-to-Speech API
+ */
+import type { SwitchCondInnerStateCurrentCanvas, SwitchCondData, SwitchDef } from '../switch'
+import type { Data, DataLink } from '../data'
+import type {
+  Brick,
+  Generator,
+  EventAction,
+  ActionWithDataParams,
+  ActionWithParams,
+  Action,
+  EventProperty,
+} from '../common'
+import type { TemplateEventPropsMap } from '../../utils/event-props'
+
+/* Generate audio */
+export type GeneratorOpenAiTTSActionGenerate = ActionWithParams & {
+  __actionName: 'GENERATOR_OPENAI_TTS_GENERATE'
+  params?: Array<{
+    input: 'text'
+    value?: string | DataLink | EventProperty
+    mapping?: string
+  }>
+}
+
+/* Clean cache */
+export type GeneratorOpenAiTTSActionCleanCache = Action & {
+  __actionName: 'GENERATOR_OPENAI_TTS_CLEAN_CACHE'
+}
+
+/* Cancel current generation and clear queue */
+export type GeneratorOpenAiTTSActionCancel = Action & {
+  __actionName: 'GENERATOR_OPENAI_TTS_CANCEL'
+}
+
+interface GeneratorOpenAiTTSDef {
+  /*
+Default property:
+{
+  "apiEndpoint": "https://api.openai.com/v1",
+  "model": "tts-1",
+  "voice": "alloy",
+  "speed": 1,
+  "outputType": "play",
+  "playbackVolume": 100,
+  "cacheGenerated": true,
+  "autoInferEnable": false,
+  "softBreakRegex": "^[^\\r\\n\\t\\f\\v]*([\\r\\n]+|[。！？!?.]\\B)",
+  "hardBreakTime": 500
+}
+  */
+  property?: {
+    /* API endpoint URL */
+    apiEndpoint?: string | DataLink
+    /* OpenAI API Key */
+    apiKey?: string | DataLink
+    /* OpenAI TTS model */
+    model?: string | DataLink
+    /* Voice to use
+ Select voice from https://openai.fm , default alloy */
+    voice?: string | DataLink
+    /* Additional instructions for the speech generation */
+    instructions?: string | DataLink
+    /* Speed of the generated audio */
+    speed?: number | DataLink
+    /* Output mode */
+    outputType?: 'play' | 'file' | DataLink
+    /* Playback volume (0 - 100) */
+    playbackVolume?: number | DataLink
+    /* Enable cache for generated audio */
+    cacheGenerated?: boolean | DataLink
+    /* Text to generate */
+    prompt?: string | DataLink
+    /* Auto inference when prompt changes */
+    autoInferEnable?: boolean | DataLink
+    /* Segmentation rule for auto inference */
+    softBreakRegex?: string | DataLink
+    /* Time to force inference when softBreakRegex is not satisfied */
+    hardBreakTime?: number | DataLink
+  }
+  events?: {
+    /* Event triggered when state change */
+    onContextStateChange?: Array<
+      EventAction<string & keyof TemplateEventPropsMap['OpenaiTts']['onContextStateChange']>
+    >
+    /* Event triggered when error occurs */
+    onError?: Array<EventAction<string & keyof TemplateEventPropsMap['OpenaiTts']['onError']>>
+  }
+  outlets?: {
+    /* Context state */
+    contextState?: () => Data<string>
+    /* Generated audio file */
+    generatedAudio?: () => Data<string>
+    /* Generated audio file is playing */
+    generatedAudioPlaying?: () => Data<boolean>
+  }
+}
+
+/* Generate speech from text using OpenAI's Text-to-Speech API */
+export type GeneratorOpenAiTTS = Generator &
+  GeneratorOpenAiTTSDef & {
+    templateKey: 'GENERATOR_OPENAI_TTS'
+    switches?: Array<
+      SwitchDef &
+        GeneratorOpenAiTTSDef & {
+          conds?: Array<{
+            method: '==' | '!=' | '>' | '<' | '>=' | '<='
+            cond:
+              | SwitchCondInnerStateCurrentCanvas
+              | SwitchCondData
+              | {
+                  __typename: 'SwitchCondInnerStateOutlet'
+                  outlet: 'contextState' | 'generatedAudio' | 'generatedAudioPlaying'
+                  value: any
+                }
+          }>
+        }
+    >
+  }