echogarden 1.0.5 → 1.1.0

package/dist/recognition/WhisperSTT.js

@@ -1,10 +1,10 @@
 import { Logger } from '../utilities/Logger.js';
 import { computeMelSpectogramUsingFilterbanks } from '../dsp/MelSpectogram.js';
-import { clip, getIntegerRange, getRepetitionScoreRelativeToFirstSubstring, getUTF32Chars, splitFloat32Array, yieldToEventLoop } from '../utilities/Utilities.js';
-import { indexOfMax, logOfVector, logSumExp, meanOfVector, medianFilter, softmax, stdDeviationOfVector } from '../math/VectorMath.js';
+import { clip, containsInvalidCodepoint, getIntegerRange, getTokenRepetitionScore, splitFloat32Array, yieldToEventLoop } from '../utilities/Utilities.js';
+import { indexOfMax, logOfVector, logSumExp, meanOfVector, softmax, stdDeviationOfVector } from '../math/VectorMath.js';
 import { alignDTWWindowed } from '../alignment/DTWSequenceAlignmentWindowed.js';
 import { extendDeep } from '../utilities/ObjectUtilities.js';
-import { getRawAudioDuration } from '../audio/AudioUtilities.js';
+import { getRawAudioDuration, sliceRawAudio } from '../audio/AudioUtilities.js';
 import { readFile } from '../utilities/FileSystem.js';
 import path from 'path';
 import { getShortLanguageCode, languageCodeToName } from '../utilities/Locale.js';
@@ -12,8 +12,12 @@ import { loadPackage } from '../utilities/PackageManager.js';
 import chalk from 'chalk';
 import { XorShift32RNG } from '../utilities/RandomGenerator.js';
 import { detectSpeechLanguageByParts } from '../api/LanguageDetection.js';
-import { isPunctuation, isWhitespace } from '../nlp/Segmentation.js';
+import { isPunctuation, isWhitespace, isWord, splitToSentences, splitToWords } from '../nlp/Segmentation.js';
+import { medianOf5Filter } from '../math/MedianFilter.js';
+import { getDeflateCompressionMetricsForString } from '../utilities/Compression.js';
+import { getOnnxSessionOptions, makeOnnxLikeFloat32Tensor } from '../utilities/OnnxUtilities.js';
 export async function recognize(sourceRawAudio, modelName, modelDir, task, sourceLanguage, options) {
+    options = extendDeep(defaultWhisperOptions, options);
     if (sourceRawAudio.sampleRate != 16000) {
         throw new Error('Source audio must have a sampling rate of 16000');
     }
@@ -31,11 +35,14 @@ export async function recognize(sourceRawAudio, modelName, modelDir, task, sourc
     if (seed) {
         seed = Math.max(Math.floor(seed), 1) | 0;
     }
-    const whisper = new Whisper(modelName, modelDir, seed);
+    const encoderProviders = options.encoderProvider ? [options.encoderProvider] : ['dml', 'cpu'];
+    const decoderProviders = options.decoderProvider ? [options.decoderProvider] : [];
+    const whisper = new Whisper(modelName, modelDir, encoderProviders, decoderProviders, seed);
     const result = await whisper.recognize(sourceRawAudio, task, sourceLanguage, options);
     return result;
 }
-export async function align(sourceRawAudio, referenceText, modelName, modelDir, sourceLanguage) {
+export async function align(sourceRawAudio, transcript, modelName, modelDir, sourceLanguage, options) {
+    options = extendDeep(defaultWhisperAlignmentOptions, options);
     if (sourceRawAudio.sampleRate != 16000) {
         throw new Error('Source audio must have a sampling rate of 16000');
     }
@@ -46,46 +53,101 @@ export async function align(sourceRawAudio, referenceText, modelName, modelDir,
     if (isEnglishOnlyModel(modelName) && sourceLanguage != 'en') {
         throw new Error(`The model '${modelName}' can only be used with English inputs. However, the given source language was ${languageCodeToName(sourceLanguage)}.`);
     }
-    const whisper = new Whisper(modelName, modelDir);
-    const timeline = await whisper.align(sourceRawAudio, referenceText, sourceLanguage);
+    const encoderProviders = options.encoderProvider ? [options.encoderProvider] : ['dml', 'cpu'];
+    const decoderProviders = options.decoderProvider ? [options.decoderProvider] : [];
+    const whisper = new Whisper(modelName, modelDir, encoderProviders, decoderProviders);
+    const timeline = await whisper.align(sourceRawAudio, transcript, sourceLanguage, 'transcribe', options);
     return timeline;
 }
-export async function detectLanguage(sourceRawAudio, modelName, modelDir, temperature) {
+export async function alignEnglishTranslation(sourceRawAudio, translatedTranscript, modelName, modelDir, sourceLanguage, options) {
+    options = extendDeep(defaultWhisperAlignmentOptions, options);
+    if (sourceRawAudio.sampleRate != 16000) {
+        throw new Error('Source audio must have a sampling rate of 16000');
+    }
+    sourceLanguage = getShortLanguageCode(sourceLanguage);
+    if (!(sourceLanguage in languageIdLookup)) {
+        throw new Error(`The source language ${languageCodeToName(sourceLanguage)} is not supported by the Whisper engine.`);
+    }
+    if (isEnglishOnlyModel(modelName)) {
+        throw new Error(`Translation alignment can only be done with multilingual models.`);
+    }
+    const encoderProviders = options.encoderProvider ? [options.encoderProvider] : ['dml', 'cpu'];
+    const decoderProviders = options.decoderProvider ? [options.decoderProvider] : [];
+    const whisper = new Whisper(modelName, modelDir, encoderProviders, decoderProviders);
+    const timeline = await whisper.align(sourceRawAudio, translatedTranscript, sourceLanguage, 'translate', options);
+    return timeline;
+}
+export async function detectLanguage(sourceRawAudio, modelName, modelDir, options) {
+    options = extendDeep(defaultWhisperLanguageDetectionOptions, options);
     if (sourceRawAudio.sampleRate != 16000) {
         throw new Error('Source audio must have a sampling rate of 16000');
     }
     if (!isMultilingualModel(modelName)) {
         throw new Error(`Language detection is only supported with multilingual models.`);
     }
-    if (temperature < 0) {
+    if (options.temperature < 0) {
         throw new Error(`Temperature cannot be negative`);
     }
-    const whisper = new Whisper(modelName, modelDir);
+    const encoderProviders = options.encoderProvider ? [options.encoderProvider] : ['dml', 'cpu'];
+    const decoderProviders = options.decoderProvider ? [options.decoderProvider] : [];
+    const whisper = new Whisper(modelName, modelDir, encoderProviders, decoderProviders);
     async function detectLanguageForPart(partAudio) {
         const audioFeatures = await whisper.encodeAudio(partAudio);
-        const partResults = await whisper.detectLanguage(audioFeatures, temperature);
+        const partResults = await whisper.detectLanguage(audioFeatures, options.temperature);
         return partResults;
     }
     const results = await detectSpeechLanguageByParts(sourceRawAudio, detectLanguageForPart);
     results.sort((entry1, entry2) => entry2.probability - entry1.probability);
     return results;
 }
+export async function detectVoiceActivity(sourceRawAudio, modelName, modelDir, options) {
+    options = extendDeep(defaultWhisperVADOptions, options);
+    if (sourceRawAudio.sampleRate != 16000) {
+        throw new Error('Source audio must have a sampling rate of 16000');
+    }
+    if (options.temperature < 0) {
+        throw new Error(`Temperature cannot be negative`);
+    }
+    const audioSamples = sourceRawAudio.audioChannels[0];
+    const partDuration = 5;
+    const maxSamplesCountForPart = sourceRawAudio.sampleRate * partDuration;
+    const encoderProviders = options.encoderProvider ? [options.encoderProvider] : ['dml', 'cpu'];
+    const decoderProviders = options.decoderProvider ? [options.decoderProvider] : [];
+    const whisper = new Whisper(modelName, modelDir, encoderProviders, decoderProviders);
+    const partProbabilities = [];
+    for (let sampleOffset = 0; sampleOffset < audioSamples.length; sampleOffset += maxSamplesCountForPart) {
+        const partSamples = sliceRawAudio(sourceRawAudio, sampleOffset, sampleOffset + maxSamplesCountForPart);
+        const samplesCountForPart = partSamples.audioChannels[0].length;
+        const startTime = sampleOffset / sourceRawAudio.sampleRate;
+        const endTime = (sampleOffset + samplesCountForPart) / sourceRawAudio.sampleRate;
+        const encodedPartSamples = await whisper.encodeAudio(partSamples);
+        const probabilityForPart = await whisper.detectVoiceActivity(encodedPartSamples, options.temperature);
+        partProbabilities.push({
+            type: 'segment',
+            text: '',
+            startTime,
+            endTime,
+            confidence: probabilityForPart,
+        });
+    }
+    return { partProbabilities };
+}
 export class Whisper {
     modelName;
     modelDir;
+    encoderExecutionProviders;
+    decoderExecutionProviders;
     isMultiligualModel;
     audioEncoder;
     textDecoder;
     tiktoken;
-    onnxOptions = {
-        logSeverityLevel: 2,
-        executionProviders: ['cpu']
-    };
     tokenConfig;
     randomGen;
-    constructor(modelName, modelDir, rngSeed = 461845907) {
+    constructor(modelName, modelDir, encoderExecutionProviders, decoderExecutionProviders, rngSeed = 461845907) {
         this.modelName = modelName;
         this.modelDir = modelDir;
+        this.encoderExecutionProviders = encoderExecutionProviders;
+        this.decoderExecutionProviders = decoderExecutionProviders;
         this.isMultiligualModel = isMultilingualModel(this.modelName);
         if (this.isMultiligualModel) {
             this.tokenConfig = {
@@ -119,87 +181,25 @@ export class Whisper {
         }
         this.randomGen = new XorShift32RNG(rngSeed);
     }
-    async initializeIfNeeded() {
-        await this.initializeTokenizerIfNeeded();
-        await this.initializeEncoderSessionIfNeeded();
-        await this.initializeDecoderSessionIfNeeded();
-    }
-    async initializeTokenizerIfNeeded() {
-        if (this.tiktoken) {
-            return;
-        }
-        const logger = new Logger();
-        await logger.startAsync('Load tokenizer data');
-        const tiktokenModulePackagePath = await loadPackage('whisper-tiktoken-data');
-        const tiktokenDataFilePath = path.join(tiktokenModulePackagePath, this.isMultiligualModel ? 'multilingual.tiktoken' : 'gpt2.tiktoken');
-        let tiktokenData = await readFile(tiktokenDataFilePath, { encoding: 'utf8' });
-        const tokenConfig = this.tokenConfig;
-        const metadataTokens = {
-            [tokenConfig.endOfTextToken]: '[EndOfText]',
-            [tokenConfig.startOfTextToken]: '[StartOfText]',
-            [tokenConfig.translateTaskToken]: '[TranslateTask]',
-            [tokenConfig.transcribeTaskToken]: '[TranscribeTask]',
-            [tokenConfig.startOfPromptToken]: '[StartOfPrompt]',
-            [tokenConfig.nonSpeechToken]: '[NonSpeech]',
-            [tokenConfig.noTimestampsToken]: '[NoTimestamps]',
-        };
-        if (this.isMultiligualModel) {
-            metadataTokens[50256] = '[Unused_50256]';
-            metadataTokens[50360] = '[Unused_50360]';
-        }
-        const languageTokenCount = tokenConfig.languageTokensEnd - tokenConfig.languageTokensStart;
-        for (let i = 0; i < languageTokenCount; i++) {
-            const tokenIndex = this.tokenConfig.languageTokensStart + i;
-            metadataTokens[tokenIndex] = `[Language_${i}]`;
-        }
-        const timestampTokensCount = 1501;
-        for (let i = 0; i < timestampTokensCount; i++) {
-            const tokenIndex = this.tokenConfig.timestampTokensStart + i;
-            const tokenTime = this.timestampTokenToSeconds(tokenIndex);
-            metadataTokens[tokenIndex] = `[Timestamp_${tokenTime.toFixed(2)}]`;
-        }
-        const inverseMetadataTokensLookup = {};
-        for (const [key, value] of Object.entries(metadataTokens)) {
-            inverseMetadataTokensLookup[value] = parseInt(key);
-        }
-        const patternString = `'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)|\\s+`;
-        const { Tiktoken } = await import('tiktoken/lite');
-        this.tiktoken = new Tiktoken(tiktokenData, inverseMetadataTokensLookup, patternString);
-        logger.end();
-    }
-    async initializeEncoderSessionIfNeeded() {
-        if (this.audioEncoder) {
-            return;
-        }
-        const logger = new Logger();
-        await logger.startAsync(`Create encoder model inference session for model '${this.modelName}'`);
-        const encoderFilePath = path.join(this.modelDir, 'encoder.onnx');
-        const Onnx = await import('onnxruntime-node');
-        this.audioEncoder = await Onnx.InferenceSession.create(encoderFilePath, this.onnxOptions);
-        logger.end();
-    }
-    async initializeDecoderSessionIfNeeded() {
-        if (this.textDecoder) {
-            return;
-        }
-        const logger = new Logger();
-        await logger.startAsync(`Create decoder model inference session for model '${this.modelName}'`);
-        const decoderFilePath = path.join(this.modelDir, 'decoder.onnx');
-        const Onnx = await import('onnxruntime-node');
-        this.textDecoder = await Onnx.InferenceSession.create(decoderFilePath, this.onnxOptions);
-        logger.end();
-    }
-    async recognize(rawAudio, task, language, options) {
+    async recognize(rawAudio, task, language, options, logitFilter) {
         await this.initializeIfNeeded();
         const logger = new Logger();
+        options = extendDeep(defaultWhisperOptions, options);
+        options.model = this.modelName;
         const audioSamples = rawAudio.audioChannels[0];
         const sampleRate = rawAudio.sampleRate;
         const prompt = options.prompt;
-        const maxAudioSamplesPerPart = sampleRate * 30;
         const decodeTimestampTokens = options.decodeTimestampTokens;
+        const maxAudioSamplesPerPart = sampleRate * 30;
         let previousPartTextTokens = [];
         let timeline = [];
         let allDecodedTokens = [];
+        let wrappedLogitFilter;
+        if (logitFilter) {
+            wrappedLogitFilter = (logits, partDecodedTokens, isFirstPart, isFinalPart) => {
+                return logitFilter(logits, [...allDecodedTokens, ...partDecodedTokens], isFirstPart, isFinalPart);
+            };
+        }
         for (let audioOffset = 0; audioOffset < audioSamples.length;) {
             const segmentStartTime = audioOffset / sampleRate;
             await logger.startAsync(`\nPrepare audio part at time position ${segmentStartTime.toFixed(2)}`, undefined, chalk.magentaBright);
@@ -220,7 +220,7 @@ export class Whisper {
             }
             initialTokens = [...initialTokens, ...this.getTextStartTokens(language, task, !decodeTimestampTokens)];
             logger.end();
-            let { decodedTokens: partTokens, crossAttentionQKs: partCrossAttentionQKs, decodedTokensConfidence } = await this.decodeTokens(audioPartFeatures, initialTokens, audioPartDuration, isFirstPart, isFinalPart, options);
+            let { decodedTokens: partTokens, decodedTokensConfidence: partTokensConfidence, decodedTokensCrossAttentionQKs: partCrossAttentionQKs, } = await this.decodeTokens(audioPartFeatures, initialTokens, audioPartDuration, isFirstPart, isFinalPart, options, wrappedLogitFilter);
             const lastToken = partTokens[partTokens.length - 1];
             const lastTokenIsTimestamp = this.isTimestampToken(lastToken);
             let audioEndOffset;
@@ -237,50 +237,96 @@ export class Whisper {
             if (partTokens.length != partCrossAttentionQKs.length) {
                 throw new Error('Unexpected: partTokens.length != partCrossAttentionQKs.length');
             }
+            // Prepare tokens
             partTokens = partTokens.slice(initialTokens.length);
-            //const compressionRatioForPart = (await getDeflateCompressionMetricsForString(this.tokensToText(partTokens))).ratio
+            partTokensConfidence = partTokensConfidence.slice(initialTokens.length);
             partCrossAttentionQKs = partCrossAttentionQKs.slice(initialTokens.length);
+            // Compute compression ratio for part
+            if (false) {
+                const compressionRatioForPart = (await getDeflateCompressionMetricsForString(this.tokensToText(partTokens))).ratio;
+            }
+            // Find alignment path
             const alignmentPath = await this.findAlignmentPathFromQKs(partCrossAttentionQKs, partTokens, 0, segmentFrameCount); //, alignmentHeadsIndexes[this.modelName])
-            const partTimeline = await this.getTokenTimelineFromAlignmentPath(alignmentPath, partTokens, segmentStartTime, segmentEndTime, decodedTokensConfidence);
-            audioOffset = audioEndOffset;
+            // Generate timeline from alignment path
+            const partTimeline = await this.getTokenTimelineFromAlignmentPath(alignmentPath, partTokens, segmentStartTime, segmentEndTime, partTokensConfidence);
             allDecodedTokens.push(...partTokens);
             timeline.push(...partTimeline);
             previousPartTextTokens = partTokens.filter(token => this.isTextToken(token));
+            audioOffset = audioEndOffset;
             logger.end();
         }
+        // Convert token timeline to word timeline
         timeline = this.tokenTimelineToWordTimeline(timeline, language);
+        // Convert tokens to transcript
         const transcript = this.tokensToText(allDecodedTokens).trim();
         logger.end();
         return { transcript, timeline };
     }
-    async align(rawAudio, referenceText, language) {
-        await this.initializeIfNeeded();
-        const logger = new Logger();
-        await logger.startAsync('Prepare for alignment');
-        referenceText = referenceText.replaceAll(/\s+/g, ' ');
-        const audioDuration = Math.min(getRawAudioDuration(rawAudio), 30);
-        const audioFrameCount = this.secondsToFrame(audioDuration);
-        const initialTokens = this.getTextStartTokens(language, 'transcribe', true);
+    async align(rawAudio, transcript, sourceLanguage, task, whisperAlignmentOptions) {
+        await this.initializeTokenizerIfNeeded();
+        whisperAlignmentOptions = extendDeep(defaultWhisperAlignmentOptions, whisperAlignmentOptions);
+        const shouldSplitToSentences = false;
+        let simplifiedTranscript = '';
+        if (shouldSplitToSentences) {
+            const sentences = splitToSentences(transcript, 'en');
+            for (const sentence of sentences) {
+                let sentenceWords = await splitToWords(sentence, 'en');
+                sentenceWords = sentenceWords.filter(word => isWord(word));
+                simplifiedTranscript += sentenceWords.join(' ');
+                simplifiedTranscript += ' ';
+            }
+        }
+        else {
+            let words = await splitToWords(transcript, 'en');
+            words = words.map(word => word.trim());
+            words = words.filter(word => isWord(word));
+            simplifiedTranscript = words.join(' ');
+        }
+        // Tokenize the transcript
+        const simplifiedTranscriptTokens = this.textToTokens(simplifiedTranscript);
+        // Initialize custom logit filter that allows only the transcript tokens to be decoded
+        // in order.
         const endOfTextToken = this.tokenConfig.endOfTextToken;
-        let tokens = [...initialTokens, ...this.textToTokens(referenceText), endOfTextToken];
-        logger.end();
-        const audioFeatures = await this.encodeAudio(rawAudio);
-        await logger.startAsync('Infer cross-attention QKs');
-        let crossAttentionQKs = await this.inferCrossAttentionQKs(tokens, audioFeatures);
-        tokens = tokens.slice(initialTokens.length, tokens.length - 1);
-        crossAttentionQKs = crossAttentionQKs.slice(initialTokens.length, crossAttentionQKs.length - 1);
-        await logger.startAsync('Extract word timeline');
-        const alignmentPath = await this.findAlignmentPathFromQKs(crossAttentionQKs, tokens, 0, audioFrameCount); //, this.getAlignmentHeadIndexes())
-        const tokenTimeline = await this.getTokenTimelineFromAlignmentPath(alignmentPath, tokens, 0, audioDuration);
-        const wordTimeline = this.tokenTimelineToWordTimeline(tokenTimeline, language);
-        logger.end();
-        return wordTimeline;
+        const logitFilter = (logits, decodedTokens, isFirstPart, isFinalPart) => {
+            const decodedTextTokens = decodedTokens.filter(token => this.isTextToken(token));
+            const nextTokenToDecode = simplifiedTranscriptTokens[decodedTextTokens.length] ?? endOfTextToken;
+            const newLogits = logits.map((logit, index) => {
+                if (index === nextTokenToDecode) {
+                    return logit;
+                }
+                // If it's the final part, the ent-of-text token logit is set to -Infinity.
+                // This will force to force all transcript tokens to be decoded even if the model doesn't
+                // recognize them.
+                if (!isFinalPart && index === endOfTextToken) {
+                    return logit;
+                }
+                return -Infinity;
+            });
+            return newLogits;
+        };
+        // Set options for alignment
+        const options = {
+            model: this.modelName,
+            temperature: 0.0,
+            prompt: undefined,
+            topCandidateCount: 1,
+            punctuationThreshold: Infinity,
+            autoPromptParts: false,
+            maxTokensPerPart: Infinity,
+            suppressRepetition: false,
+            decodeTimestampTokens: true,
+            endTokenThreshold: whisperAlignmentOptions.endTokenThreshold,
+            includeEndTokenInCandidates: false,
+            seed: undefined,
+        };
+        // Recognize
+        const { timeline } = await this.recognize(rawAudio, task, sourceLanguage, options, logitFilter);
+        return timeline;
     }
     async detectLanguage(audioFeatures, temperature) {
         if (!this.isMultiligualModel) {
             throw new Error('Language detection is only supported with multilingual models');
         }
-        await this.initializeTokenizerIfNeeded();
         await this.initializeDecoderSessionIfNeeded();
         // Prepare and run decoder
         const logger = new Logger();
@@ -317,38 +363,76 @@ export class Whisper {
         logger.end();
         return results;
     }
-    async decodeTokens(audioFeatures, initialTokens, audioDuration, isFirstPart, isFinalPart, options) {
+    async detectVoiceActivity(audioFeatures, temperature) {
+        await this.initializeDecoderSessionIfNeeded();
+        // Prepare and run decoder
+        const logger = new Logger();
+        await logger.startAsync('Detect voice activity with Whisper model');
+        const sotToken = this.tokenConfig.startOfTextToken;
+        const initialTokens = [sotToken];
+        const offset = 0;
+        const Onnx = await import('onnxruntime-node');
+        const initialKvDimensions = this.getKvDimensions(1, initialTokens.length);
+        const kvCacheTensor = new Onnx.Tensor('float32', new Float32Array(initialKvDimensions[0] * initialKvDimensions[1] * initialKvDimensions[2] * initialKvDimensions[3]), initialKvDimensions);
+        const tokensTensor = new Onnx.Tensor('int64', new BigInt64Array(initialTokens.map(token => BigInt(token))), [1, initialTokens.length]);
+        const offsetTensor = new Onnx.Tensor('int64', new BigInt64Array([BigInt(offset)]), []);
+        const decoderInputs = {
+            tokens: tokensTensor,
+            audio_features: audioFeatures,
+            kv_cache: kvCacheTensor,
+            offset: offsetTensor
+        };
+        const decoderOutputs = await this.textDecoder.run(decoderInputs);
+        const logitsBuffer = decoderOutputs['logits'].data;
+        const tokenConfig = this.tokenConfig;
+        const logits = Array.from(logitsBuffer);
+        const probabilities = softmax(logits, temperature);
+        const noSpeechProbability = probabilities[tokenConfig.nonSpeechToken];
+        return 1.0 - noSpeechProbability;
+    }
+    // Decode tokens using the decoder model
+    async decodeTokens(audioFeatures, initialTokens, audioDuration, isFirstPart, isFinalPart, options, logitFilter) {
+        // Initialize
         await this.initializeTokenizerIfNeeded();
         await this.initializeDecoderSessionIfNeeded();
         const logger = new Logger();
-        const allowedPunctuationMarks = this.getAllowedPunctuationMarks();
         await logger.startAsync('Decode text tokens with Whisper decoder model');
         options = extendDeep(defaultWhisperOptions, options);
         const Onnx = await import('onnxruntime-node');
+        // Get token information
         const endOfTextToken = this.tokenConfig.endOfTextToken;
         const timestampTokensStart = this.tokenConfig.timestampTokensStart;
-        const suppressedTokens = new Set(this.getSuppressedTokens());
+        const suppressedTextTokens = this.getSuppressedTextTokens();
+        const suppressedMetadataTokens = this.getSuppressedMetadataTokens();
+        const allowedPunctuationMarks = this.getAllowedPunctuationMarks();
         const spaceToken = this.textToTokens(' ')[0];
-        const maxDecodedTokenCount = options.maxTokensPerPart;
+        // Initialize variables for decoding loop
         let decodedTokens = initialTokens.slice();
         const initialKvDimensions = this.getKvDimensions(1, decodedTokens.length);
         let kvCacheTensor = new Onnx.Tensor('float32', new Float32Array(initialKvDimensions[0] * initialKvDimensions[1] * initialKvDimensions[2] * initialKvDimensions[3]), initialKvDimensions);
-        let decodedTokensTimestampLogits = [new Array(1501)];
-        let lastTimestampTokenIndex = -1;
-        let timestampsSeenCount = 0;
-        const decodedTokensConfidence = [];
+        let decodedTokensTimestampLogits = [];
+        let decodedTokensConfidence = [];
         let decodedTokensCrossAttentionQKs = [];
         for (let i = 0; i < decodedTokens.length; i++) {
+            decodedTokensTimestampLogits.push(new Array(1501));
+            decodedTokensConfidence.push(1.0);
             decodedTokensCrossAttentionQKs.push(undefined);
         }
+        let lastTimestampTokenIndex = -1;
+        let timestampTokenSeenCount = 0;
         let bufferedTokensToPrint = [];
+        // Define method to add a token to output
+        function addToken(tokenToAdd, timestampLogits, confidence, crossAttentionQKs) {
+            decodedTokens.push(tokenToAdd);
+            decodedTokensTimestampLogits.push(timestampLogits);
+            decodedTokensConfidence.push(confidence);
+            decodedTokensCrossAttentionQKs.push(crossAttentionQKs);
+        }
         // Start decoding loop
-        for (let decodedTokenCount = 0; decodedTokenCount < maxDecodedTokenCount; decodedTokenCount++) {
+        for (let decodedTokenCount = 0; decodedTokenCount < options.maxTokensPerPart; decodedTokenCount++) {
             const isInitialState = decodedTokens.length == initialTokens.length;
-            const tokensToDecode = isInitialState ? decodedTokens : [decodedTokens[decodedTokens.length - 1]];
-            const offset = isInitialState ? 0 : decodedTokens.length;
+            // If not in initial state, reshape KV Cache tensor to accomodate a new output token
             if (!isInitialState) {
-                // Reshape KV Cache tensor
                 const dims = kvCacheTensor.dims;
                 const currentKvCacheGroups = splitFloat32Array(kvCacheTensor.data, dims[2] * dims[3]);
                 const reshapedKvCacheTensor = new Onnx.Tensor('float32', new Float32Array(dims[0] * dims[1] * (decodedTokens.length) * dims[3]), [dims[0], dims[1], decodedTokens.length, dims[3]]);
@@ -358,199 +442,252 @@ export class Whisper {
                 }
                 kvCacheTensor = reshapedKvCacheTensor;
             }
-            // Prepare and run decoder
+            // Prepare values for decoder
+            const tokensToDecode = isInitialState ? decodedTokens : [decodedTokens[decodedTokens.length - 1]];
+            const offset = isInitialState ? 0 : decodedTokens.length;
             const tokensTensor = new Onnx.Tensor('int64', new BigInt64Array(tokensToDecode.map(token => BigInt(token))), [1, tokensToDecode.length]);
             const offsetTensor = new Onnx.Tensor('int64', new BigInt64Array([BigInt(offset)]), []);
-            const decoderInputs = { tokens: tokensTensor, audio_features: audioFeatures, kv_cache: kvCacheTensor, offset: offsetTensor };
+            const decoderInputs = {
+                tokens: tokensTensor,
+                audio_features: audioFeatures,
+                kv_cache: kvCacheTensor,
+                offset: offsetTensor
+            };
+            // Run decoder
             const decoderOutputs = await this.textDecoder.run(decoderInputs);
+            // Store results
             const logitsBuffer = decoderOutputs['logits'].data;
             kvCacheTensor = decoderOutputs['output_kv_cache'];
-            // Compute logits
-            const resultLogits = splitFloat32Array(logitsBuffer, logitsBuffer.length / decoderOutputs['logits'].dims[1]);
-            const allTokenLogits = Array.from(resultLogits[resultLogits.length - 1]);
-            const timestampTokenLogits = allTokenLogits.slice(timestampTokensStart);
-            // Suppress logits for tokens in the suppressed set
-            for (let logitIndex = 0; logitIndex < allTokenLogits.length; logitIndex++) {
-                const isWrongTokenForInitialState = isInitialState &&
-                    (logitIndex === spaceToken || logitIndex === endOfTextToken);
-                const isInSuppressedList = suppressedTokens.has(logitIndex);
-                const shouldSuppressToken = isWrongTokenForInitialState || isInSuppressedList;
-                if (shouldSuppressToken) {
-                    allTokenLogits[logitIndex] = -Infinity;
-                }
+            const crossAttentionQKsForTokenOnnx = decoderOutputs['cross_attention_qks'];
+            const crossAttentionQKsForToken = makeOnnxLikeFloat32Tensor(crossAttentionQKsForTokenOnnx);
+            crossAttentionQKsForTokenOnnx.dispose();
+            // Get logits
+            const resultLogitsFloatArrays = splitFloat32Array(logitsBuffer, logitsBuffer.length / decoderOutputs['logits'].dims[1]);
+            const allTokenLogits = Array.from(resultLogitsFloatArrays[resultLogitsFloatArrays.length - 1]);
+            // Suppress metadata tokens in the suppression set
+            for (const suppressedTokenIndex of suppressedMetadataTokens) {
+                allTokenLogits[suppressedTokenIndex] = -Infinity;
             }
-            // Add best token
-            function addToken(tokenToAdd, timestampLogits, confidence) {
-                decodedTokens.push(tokenToAdd);
-                decodedTokensTimestampLogits.push(timestampLogits);
-                decodedTokensCrossAttentionQKs.push(decoderOutputs['cross_attention_qks']);
-                decodedTokensConfidence.push(confidence);
+            if (isInitialState) {
+                // If in initial state, suppress end-of-text token
+                allTokenLogits[endOfTextToken] = -Infinity;
             }
-            let shouldDecodeNonTimestampToken = true;
-            if (options.decodeTimestampTokens) {
+            const timestampTokenLogits = allTokenLogits.slice(timestampTokensStart);
+            const decodeTimestampTokenIfNeeded = () => {
+                // Try to decode a timestamp token, if needed
+                // If timestamp tokens is disabled in options, don't decode a timestamp
+                if (!options.decodeTimestampTokens) {
+                    return false;
+                }
+                const previousTokenWasTimestamp = this.isTimestampToken(decodedTokens[decodedTokens.length - 1]);
+                const secondPreviousTokenWasTimestamp = this.isTimestampToken(decodedTokens[decodedTokens.length - 2]);
+                // If there are two successive timestamp tokens decoded, or the previous timestamp was the first token,
+                // don't decode a timestamp
+                if (previousTokenWasTimestamp &&
+                    (decodedTokens.length === initialTokens.length + 1) || secondPreviousTokenWasTimestamp) {
+                    return false;
+                }
                 // Derive token probabilities
                 const probabilities = softmax(allTokenLogits, 1.0);
                 const logProbabilities = logOfVector(probabilities);
                 const nonTimestampTokenLogProbs = logProbabilities.slice(0, timestampTokensStart);
+                // Find highest non-timestamp token
                 const indexOfMaxNonTimestampLogProb = indexOfMax(nonTimestampTokenLogProbs);
                 const valueOfMaxNonTimestampLogProb = nonTimestampTokenLogProbs[indexOfMaxNonTimestampLogProb];
+                // Find highest timestamp token
                 const timestampTokenLogProbs = logProbabilities.slice(timestampTokensStart);
                 const indexOfMaxTimestampLogProb = indexOfMax(timestampTokenLogProbs);
+                // Compute the log of the sum of exponentials of the log probabilities
+                // of the timestamp tokens
                 const logSumExpOfTimestampTokenLogProbs = logSumExp(timestampTokenLogProbs);
-                const shouldDecodeTimestampToken = logSumExpOfTimestampTokenLogProbs > valueOfMaxNonTimestampLogProb;
-                const previousTokenWasTimestamp = this.isTimestampToken(decodedTokens[decodedTokens.length - 1]);
-                const secondPreviousTokenWasTimestamp = decodedTokens.length < 2 || this.isTimestampToken(decodedTokens[decodedTokens.length - 2]);
-                if (shouldDecodeTimestampToken && !previousTokenWasTimestamp) {
-                    timestampsSeenCount += 1;
+                // If the sum isn't greater than the log probability of the highest non-timestamp token,
+                // don't decode a timestamp
+                if (logSumExpOfTimestampTokenLogProbs <= valueOfMaxNonTimestampLogProb) {
+                    return false;
                 }
-                if (shouldDecodeTimestampToken || (previousTokenWasTimestamp && !secondPreviousTokenWasTimestamp)) {
-                    if (previousTokenWasTimestamp) {
-                        const previousToken = decodedTokens[decodedTokens.length - 1];
-                        const previousTokenTimestampLogits = decodedTokensTimestampLogits[decodedTokensTimestampLogits.length - 1];
-                        const previousTokenConfidence = decodedTokensConfidence[decodedTokensConfidence.length - 1];
-                        addToken(previousToken, previousTokenTimestampLogits, previousTokenConfidence);
-                        lastTimestampTokenIndex = decodedTokens.length;
-                        const previousTokenTimestamp = this.timestampTokenToSeconds(previousToken);
-                        if (previousTokenTimestamp >= audioDuration) {
-                            break;
-                        }
-                    }
-                    else {
-                        const timestampToken = timestampTokensStart + indexOfMaxTimestampLogProb;
-                        const confidence = probabilities[timestampToken];
-                        addToken(timestampToken, timestampTokenLogits, confidence);
-                    }
-                    shouldDecodeNonTimestampToken = false;
+                // Decode a timestamp token
+                timestampTokenSeenCount += 1;
+                if (previousTokenWasTimestamp) {
+                    const previousToken = decodedTokens[decodedTokens.length - 1];
+                    const previousTokenTimestampLogits = decodedTokensTimestampLogits[decodedTokensTimestampLogits.length - 1];
+                    const previousTokenConfidence = decodedTokensConfidence[decodedTokensConfidence.length - 1];
+                    addToken(previousToken, previousTokenTimestampLogits, previousTokenConfidence, crossAttentionQKsForToken);
+                    lastTimestampTokenIndex = decodedTokens.length;
+                }
+                else {
+                    const timestampToken = timestampTokensStart + indexOfMaxTimestampLogProb;
+                    const confidence = probabilities[timestampToken];
+                    addToken(timestampToken, timestampTokenLogits, confidence, crossAttentionQKsForToken);
                 }
+                return true;
+            };
+            const timestampTokenDecoded = decodeTimestampTokenIfNeeded();
+            if (timestampTokenDecoded) {
+                await yieldToEventLoop();
+                continue;
             }
-            if (shouldDecodeNonTimestampToken) {
-                const topLogitCount = options.topCandidateCount;
-                const nonTimestampTokenLogits = allTokenLogits.slice(0, timestampTokensStart);
-                const sortedNonTimestampTokenLogitsWithIndexes = Array.from(nonTimestampTokenLogits).map((logit, index) => ({ token: index, logit }));
-                sortedNonTimestampTokenLogitsWithIndexes.sort((a, b) => b.logit - a.logit);
-                let topCandidates = sortedNonTimestampTokenLogitsWithIndexes.slice(0, topLogitCount)
-                    .map(entry => ({
-                    token: entry.token,
-                    logit: entry.logit,
-                    text: this.tokenToText(entry.token, true)
-                }));
-                //// Repetition suppression code
-                if (options.suppressRepetition) {
-                    const topCandidatesRepetitionScores = topCandidates.map(entry => {
-                        const lastDecodedTextTokens = decodedTokens.filter(token => this.isTextToken(token)).reverse().slice(0, 20);
-                        const { maxScore } = getRepetitionScoreRelativeToFirstSubstring([entry.token, ...lastDecodedTextTokens]);
-                        return maxScore;
-                    });
-                    const thresholdRepetitionScore = 4;
-                    if (topCandidatesRepetitionScores.every(score => score >= thresholdRepetitionScore)) {
-                        const indexOfMaxScore = topCandidatesRepetitionScores.indexOf(Math.max(...topCandidatesRepetitionScores));
-                        topCandidates = [topCandidates[indexOfMaxScore]];
-                    }
-                    else {
-                        topCandidates = topCandidates.filter((candidate, index) => topCandidatesRepetitionScores[index] < thresholdRepetitionScore);
-                    }
+            // Decode a non-timestamp token
+            let nonTimestampTokenLogits = allTokenLogits.slice(0, timestampTokensStart);
+            let shouldDecodeEndfOfTextToken = false;
+            // If not in initial state, and the end-of-text token's probability is sufficiently higher than
+            // the second highest ranked token, then set to accept it
+            if (!isInitialState) {
+                const endOfTextTokenLogit = nonTimestampTokenLogits[endOfTextToken];
+                const otherTokensLogits = nonTimestampTokenLogits.slice();
+                otherTokensLogits[endOfTextToken] = -Infinity;
+                const indexOfMaximumOtherTokenLogit = indexOfMax(otherTokensLogits);
+                const maximumOtherTokenLogit = nonTimestampTokenLogits[indexOfMaximumOtherTokenLogit];
+                const endProbabilities = softmax([endOfTextTokenLogit, maximumOtherTokenLogit], 1.0);
+                if (endProbabilities[0] > options.endTokenThreshold) {
+                    shouldDecodeEndfOfTextToken = true;
                 }
-                ////
-                const topCandidateProbabilities = softmax(topCandidates.map(a => a.logit), options.temperature);
-                //// Remove end-of-text token from candidates if its probability isn't high enough
-                if (options.decodeTimestampTokens === false) {
-                    topCandidates = topCandidates.filter((candidate, index) => {
-                        if (candidate.token === endOfTextToken) {
-                            return topCandidateProbabilities[index] >= 0.9;
-                        }
-                        return true;
-                    });
+            }
+            if (logitFilter) {
+                // Apply custom logit filter function if given
+                nonTimestampTokenLogits = logitFilter(nonTimestampTokenLogits, decodedTokens, isFirstPart, isFinalPart);
+                // If the custom filter set the end-of-text token to be Infinity, or -Infinity,
+                // then override any previous decision and accept or reject it, respectively
+                if (nonTimestampTokenLogits[endOfTextToken] === Infinity) {
+                    shouldDecodeEndfOfTextToken = true;
                 }
-                ////
-                const rankOfPromisingPunctuationToken = topCandidates.findIndex((entry, index) => {
-                    const tokenText = this.tokenToText(entry.token).trim();
-                    const isPunctuationToken = allowedPunctuationMarks.includes(tokenText);
-                    if (!isPunctuationToken) {
-                        return false;
-                    }
-                    const tokenProb = topCandidateProbabilities[index];
-                    return tokenProb >= options.punctuationThreshold;
-                });
-                let rankOfSpaceToken = topCandidates.findIndex(candidate => candidate.token === spaceToken);
-                if (rankOfSpaceToken < 0) {
-                    rankOfSpaceToken = Infinity;
+                else if (nonTimestampTokenLogits[endOfTextToken] === -Infinity) {
+                    shouldDecodeEndfOfTextToken = false;
                 }
-                let chosenCandidateRank;
-                if (rankOfPromisingPunctuationToken >= 0 &&
-                    rankOfPromisingPunctuationToken < rankOfSpaceToken) {
-                    chosenCandidateRank = rankOfPromisingPunctuationToken;
+                // If filter caused all word token logits to be -Infinity, then there is no
+                // other token to decode. Fall back to accept end-of-text
+                if (nonTimestampTokenLogits.slice(0, endOfTextToken).every(logit => logit === -Infinity)) {
+                    shouldDecodeEndfOfTextToken = true;
                 }
-                else {
-                    chosenCandidateRank = this.randomGen.selectRandomIndexFromDistribution(topCandidateProbabilities);
+            }
+            else {
+                // Otherwise, suppress text tokens in the suppression set
+                for (const suppressedTokenIndex of suppressedTextTokens) {
+                    nonTimestampTokenLogits[suppressedTokenIndex] = -Infinity;
                 }
-                const chosenToken = topCandidates[chosenCandidateRank].token;
-                const chosenTokenConfidence = topCandidateProbabilities[chosenCandidateRank];
-                addToken(chosenToken, timestampTokenLogits, chosenTokenConfidence);
-                if (chosenToken === endOfTextToken) {
-                    break;
+                // Suppress space token if at initial state
+                if (isInitialState) {
+                    nonTimestampTokenLogits[spaceToken] = -Infinity;
                 }
-                if (this.isTextToken(chosenToken)) {
-                    bufferedTokensToPrint.push(chosenToken);
-                    let textToPrint = this.tokensToText(bufferedTokensToPrint);
-                    if (textToPrint.codePointAt(0) !== 65533) {
-                        if (isFirstPart && decodedTokens.every(token => this.isMetadataToken(token))) {
-                            textToPrint = textToPrint.trimStart();
-                        }
-                        logger.write(textToPrint);
-                        bufferedTokensToPrint = [];
+            }
+            // If end-of-text token should be decoded, then add it and break
+            // out of the loop
+            if (shouldDecodeEndfOfTextToken) {
+                addToken(endOfTextToken, timestampTokenLogits, 1.0, crossAttentionQKsForToken);
+                break;
+            }
+            // Suppress end-of-text if it shouldn't be included in candidates
+            if (!options.includeEndTokenInCandidates) {
+                nonTimestampTokenLogits[endOfTextToken] = -Infinity;
+            }
+            // Find top candidates
+            const sortedNonTimestampLogitsWithIndexes = Array.from(nonTimestampTokenLogits).map((logit, index) => ({ token: index, logit }));
+            sortedNonTimestampLogitsWithIndexes.sort((a, b) => b.logit - a.logit);
+            let topCandidates = sortedNonTimestampLogitsWithIndexes.slice(0, options.topCandidateCount)
+                .map(entry => ({
+                token: entry.token,
+                logit: entry.logit,
+                text: this.tokenToText(entry.token, true)
+            }));
+            // Apply repetition suppression if enabled
+            if (options.suppressRepetition) {
+                // Using some hardcoded constants, for now
+                const tokenWindowSize = 30;
+                const thresholdMatchLength = 6;
+                const thresholdCycleRepetition = 2.0;
+                const filteredCandidates = [];
+                for (const candidate of topCandidates) {
+                    const lastDecodedTextTokens = decodedTokens
+                        .filter(token => this.isTextToken(token))
+                        .reverse()
+                        .slice(0, tokenWindowSize);
+                    const { longestMatch, longestCycleRepetition } = getTokenRepetitionScore([candidate.token, ...lastDecodedTextTokens]);
+                    if (longestMatch >= thresholdMatchLength || longestCycleRepetition >= thresholdCycleRepetition) {
+                        continue;
                     }
+                    filteredCandidates.push(candidate);
+                }
+                // If all candidates have been filtered out, accept an end-of-text token
+                if (filteredCandidates.length === 0) {
+                    filteredCandidates.push({
+                        token: endOfTextToken,
+                        logit: Infinity,
+                        text: this.tokenToText(endOfTextToken, true)
+                    });
+                }
+                topCandidates = filteredCandidates;
+            }
+            // Compute top candidate probabilities
+            const topCandidateProbabilities = softmax(topCandidates.map(a => a.logit), options.temperature);
+            // Find highest ranking punctuation token
+            const rankOfPromisingPunctuationToken = topCandidates.findIndex((entry, index) => {
+                const tokenText = this.tokenToText(entry.token).trim();
+                const isPunctuationToken = allowedPunctuationMarks.includes(tokenText);
+                if (!isPunctuationToken) {
+                    return false;
+                }
+                const tokenProb = topCandidateProbabilities[index];
+                return tokenProb >= options.punctuationThreshold;
+            });
+            // Find rank of space token
+            let rankOfSpaceToken = topCandidates.findIndex(candidate => candidate.token === spaceToken);
+            if (rankOfSpaceToken < 0) {
+                rankOfSpaceToken = Infinity;
+            }
+            // Choose token
+            let chosenCandidateRank;
+            // Select a high-ranking punctuation token if found, and it has
+            // a rank higher than the space token,
+            if (rankOfPromisingPunctuationToken >= 0 &&
+                rankOfPromisingPunctuationToken < rankOfSpaceToken) {
+                chosenCandidateRank = rankOfPromisingPunctuationToken;
+            }
+            else {
+                // Otherwise, select randomly from top k distribution
+                chosenCandidateRank = this.randomGen.selectRandomIndexFromDistribution(topCandidateProbabilities);
+            }
+            // Add chosen token
+            const chosenToken = topCandidates[chosenCandidateRank].token;
+            const chosenTokenConfidence = topCandidateProbabilities[chosenCandidateRank];
+            addToken(chosenToken, timestampTokenLogits, chosenTokenConfidence, crossAttentionQKsForToken);
+            // If chosen token is the end-of-text token, break
+            if (chosenToken === endOfTextToken) {
+                break;
+            }
+            // Print token if needed
+            if (this.isTextToken(chosenToken)) {
+                bufferedTokensToPrint.push(chosenToken);
+                let textToPrint = this.tokensToText(bufferedTokensToPrint);
+                // If the decoded text is valid, print it
+                if (!containsInvalidCodepoint(textToPrint)) {
+                    if (isFirstPart && decodedTokens.every(token => this.isMetadataToken(token))) {
+                        textToPrint = textToPrint.trimStart();
+                    }
+                    logger.write(textToPrint);
+                    bufferedTokensToPrint = [];
                 }
             }
             await yieldToEventLoop();
         }
-        if (timestampsSeenCount >= 2 && !isFinalPart) {
-            decodedTokens = decodedTokens.slice(0, lastTimestampTokenIndex);
-            decodedTokensTimestampLogits = decodedTokensTimestampLogits.slice(0, lastTimestampTokenIndex);
-            decodedTokensCrossAttentionQKs = decodedTokensCrossAttentionQKs.slice(0, lastTimestampTokenIndex);
+        // If at least two timestamp tokens were decoded and it's not the final part,
+        // truncate up to the last timestamp token
+        if (timestampTokenSeenCount >= 2 && !isFinalPart) {
+            const sliceEndTokenIndex = lastTimestampTokenIndex;
+            decodedTokens = decodedTokens.slice(0, sliceEndTokenIndex);
+            decodedTokensTimestampLogits = decodedTokensTimestampLogits.slice(0, sliceEndTokenIndex);
+            decodedTokensCrossAttentionQKs = decodedTokensCrossAttentionQKs.slice(0, sliceEndTokenIndex);
+            decodedTokensConfidence = decodedTokensConfidence.slice(0, sliceEndTokenIndex);
         }
         logger.write('\n');
         logger.end();
-        // Return the tokens
+        // Return the decoded tokens
         return {
             decodedTokens,
             decodedTokensTimestampLogits,
-            crossAttentionQKs: decodedTokensCrossAttentionQKs,
-            decodedTokensConfidence
+            decodedTokensConfidence,
+            decodedTokensCrossAttentionQKs,
         };
     }
-    async inferCrossAttentionQKs(tokens, audioFeatures) {
-        const offset = 0;
-        const Onnx = await import('onnxruntime-node');
-        const tokensTensor = new Onnx.Tensor('int64', new BigInt64Array(tokens.map(token => BigInt(token))), [1, tokens.length]);
-        const offsetTensor = new Onnx.Tensor('int64', new BigInt64Array([BigInt(offset)]), []);
-        const initialKvDimensions = this.getKvDimensions(1, tokens.length);
-        const kvCacheTensor = new Onnx.Tensor('float32', new Float32Array(initialKvDimensions[0] * initialKvDimensions[1] * initialKvDimensions[2] * initialKvDimensions[3]), initialKvDimensions);
-        const decoderInputs = { tokens: tokensTensor, audio_features: audioFeatures, kv_cache: kvCacheTensor, offset: offsetTensor };
-        const decoderOutputs = await this.textDecoder.run(decoderInputs);
-        const crossAttentionQKsTensor = decoderOutputs['cross_attention_qks'];
-        const tensorShape = crossAttentionQKsTensor.dims.slice();
-        const ndarray = (await import('ndarray')).default;
-        let qkArray = ndarray(crossAttentionQKsTensor.data, crossAttentionQKsTensor.dims.slice());
-        qkArray = qkArray.transpose(3, 0, 1, 2, 4);
-        const tokenCrossAttentionQKsTensors = [];
-        for (let i0 = 0; i0 < qkArray.shape[0]; i0++) {
-            const dataForToken = [];
-            for (let i1 = 0; i1 < qkArray.shape[1]; i1++) {
-                for (let i2 = 0; i2 < qkArray.shape[2]; i2++) {
-                    for (let i3 = 0; i3 < qkArray.shape[3]; i3++) {
-                        for (let i4 = 0; i4 < qkArray.shape[4]; i4++) {
-                            dataForToken.push(qkArray.get(i0, i1, i2, i3, i4));
-                        }
-                    }
-                }
-            }
-            const newTensorShape = tensorShape.slice();
-            newTensorShape[3] = 1;
-            const newTensor = new Onnx.Tensor('float32', dataForToken, newTensorShape);
-            tokenCrossAttentionQKsTensors.push(newTensor);
-        }
-        return tokenCrossAttentionQKsTensors;
-    }
+    // Encode audio using the encoder model
     async encodeAudio(rawAudio) {
         await this.initializeEncoderSessionIfNeeded();
         const Onnx = await import('onnxruntime-node');
@@ -562,9 +699,12 @@ export class Whisper {
         const filterbankCount = 80;
         const maxAudioSamples = sampleRate * 30;
         const maxAudioFrames = 3000;
+        if (audioSamples.length > maxAudioSamples) {
+            throw new Error(`Audio part is longer than 30 seconds`);
+        }
         await logger.startAsync('Extract mel spectogram from audio part');
         const paddedAudioSamples = new Float32Array(maxAudioSamples);
-        paddedAudioSamples.set(audioSamples.subarray(0, maxAudioSamples), 0);
+        paddedAudioSamples.set(audioSamples, 0);
         const rawAudioPart = { audioChannels: [paddedAudioSamples], sampleRate };
         const { melSpectogram } = await computeMelSpectogramUsingFilterbanks(rawAudioPart, fftOrder, fftOrder, hopLength, filterbanks);
         await logger.startAsync('Normalize mel spectogram');
@@ -592,99 +732,6 @@ export class Whisper {
         logger.end();
         return encodedAudioFeatures;
     }
-    addSegmentsToTimeline(timeline, tokens, initialTimeOffset, audioDuration) {
-        const timestampTokensStart = this.tokenConfig.timestampTokensStart;
-        for (let i = 0; i < tokens.length; i++) {
-            const token = tokens[i];
-            if (token == this.tokenConfig.startOfTextToken || token == this.tokenConfig.endOfTextToken) {
-                continue;
-            }
-            const tokenIsTimestamp = token >= timestampTokensStart;
-            const previousTokenWasTimestamp = tokens.length > 1 && tokens[i - 1] >= timestampTokensStart;
-            if (tokenIsTimestamp) {
-                if (previousTokenWasTimestamp) {
-                    continue;
-                }
-                let startTime = initialTimeOffset + this.timestampTokenToSeconds(token);
-                startTime = Math.min(startTime, audioDuration);
-                if (timeline.length > 0) {
-                    timeline[timeline.length - 1].endTime = startTime;
-                }
-                timeline.push({
-                    type: 'segment',
-                    text: '',
-                    startTime,
-                    endTime: -1,
-                });
-            }
-            else {
-                if (timeline.length == 0) {
-                    timeline.push({
-                        type: 'segment',
-                        text: '',
-                        startTime: initialTimeOffset,
-                        endTime: -1,
-                    });
-                }
-                const tokenText = this.tokenToText(token);
-                timeline[timeline.length - 1].text += tokenText;
-            }
-        }
-    }
-    async addWordsToTimeline(timeline, tokens, rawAudio, crossAttentionQKs, initialAudioTimeOffset, duration) {
-        let segmentStartTime = 0;
-        let segmentTokens = [];
-        let segmentCrossAttentionQKs = [];
-        for (let tokenIndex = 0; tokenIndex < tokens.length; tokenIndex++) {
-            const token = tokens[tokenIndex];
-            const tokenCrossAttentionQKs = crossAttentionQKs[tokenIndex];
-            const segmentTokensWithoutTimestamps = segmentTokens.filter(token => this.isNonTimestampToken(token));
-            const isTimestamp = this.isTimestampToken(token);
-            if (isTimestamp || tokenIndex == tokens.length - 1) {
-                let tokenTime;
-                if (isTimestamp) {
-                    tokenTime = this.timestampTokenToSeconds(token);
-                }
-                else {
-                    tokenTime = duration;
-                }
-                if (segmentTokensWithoutTimestamps.length > 0) {
-                    const segmentEndTime = tokenTime;
-                    const segmentStartFrame = this.secondsToFrame(segmentStartTime);
-                    let segmentEndFrame = this.secondsToFrame(segmentEndTime);
-                    if (segmentStartFrame == segmentEndFrame) {
-                        segmentEndFrame += 1;
-                    }
-                    const segmentFrameCount = segmentEndFrame - segmentStartFrame;
-                    const reinferCrossAttentionQKs = true;
-                    if (reinferCrossAttentionQKs) {
-                        const initialTokens = this.getTextStartTokens('en', 'transcribe');
-                        const tokensToDecode = [...initialTokens, ...segmentTokensWithoutTimestamps];
-                        //const segmentAudioFeaturesBuffer = audioFeatures.data.slice(segmentStartFrame * audioFeatures.dims[2], segmentEndFrame * audioFeatures.dims[2])
-                        //const segmentAudioFeatures = new Onnx.Tensor('float32', segmentAudioFeaturesBuffer, [1, segmentFrameCount, audioFeatures.dims[2]])
-                        const segmentAudioSamples = rawAudio.audioChannels[0].slice(Math.floor(segmentStartTime * rawAudio.sampleRate), Math.floor(segmentEndTime * rawAudio.sampleRate));
-                        const segmentRawAudio = { audioChannels: [segmentAudioSamples], sampleRate: rawAudio.sampleRate };
-                        const segmentAudioFeatures = await this.encodeAudio(segmentRawAudio);
-                        const reinferredCrossAttentionQKs = await this.inferCrossAttentionQKs(tokensToDecode, segmentAudioFeatures);
-                        reinferredCrossAttentionQKs.slice(initialTokens.length);
-                        const alignmentPath = await this.findAlignmentPathFromQKs(reinferredCrossAttentionQKs, tokensToDecode, 0, segmentFrameCount); //, alignmentHeadsIndexes[modelName])
-                        const tokenTimeline = await this.getTokenTimelineFromAlignmentPath(alignmentPath, segmentTokensWithoutTimestamps, initialAudioTimeOffset + segmentStartTime, initialAudioTimeOffset + segmentEndTime);
-                        timeline.push(...tokenTimeline);
-                    }
-                    else {
-                        const alignmentPath = await this.findAlignmentPathFromQKs(segmentCrossAttentionQKs, segmentTokens, segmentStartFrame, segmentEndFrame); //, alignmentHeadsIndexes[modelName])
-                        const tokenTimeline = await this.getTokenTimelineFromAlignmentPath(alignmentPath, segmentTokens, initialAudioTimeOffset, initialAudioTimeOffset + segmentEndTime);
-                        timeline.push(...tokenTimeline);
-                    }
-                }
-                segmentStartTime = tokenTime;
-                segmentTokens = [];
-                segmentCrossAttentionQKs = [];
-            }
-            segmentTokens.push(token);
-            segmentCrossAttentionQKs.push(tokenCrossAttentionQKs);
-        }
-    }
     tokenTimelineToWordTimeline(tokenTimeline, language) {
         function isSeparatorCharacter(char) {
             const nonSeparatingPunctuation = [`'`, `-`, `.`, `·`, `•`];
@@ -699,6 +746,9 @@ export class Whisper {
         function endsWithSeparatorCharacter(text) {
             return isSeparatorCharacter(text[text.length - 1]);
         }
+        if (language != 'zh' && language != 'ja') {
+            tokenTimeline = tokenTimeline.filter(entry => this.isTextToken(entry.id));
+        }
         const resultTimeline = [];
         let groups = [];
         for (let tokenIndex = 0; tokenIndex < tokenTimeline.length; tokenIndex++) {
@@ -821,35 +871,35 @@ export class Whisper {
         const applySoftmax = true;
         const normalize = true;
         const applyMedianFilter = true;
-        const fixateTimestampTokens = false;
+        const anchorTimestampTokens = false;
         const softmaxTemperature = 1.0;
-        const medianFilterWidth = 7;
+        // Apply softmax to each token's frames, if enabled
         if (applySoftmax) {
-            // Apply softmax to each token's frames
             for (const head of attentionHeads) {
                 for (let tokenIndex = 0; tokenIndex < tokenCount; tokenIndex++) {
                     head[tokenIndex] = softmax(head[tokenIndex], softmaxTemperature);
                 }
             }
         }
+        // Normalize all weights in each individual head, if enabled
         if (normalize) {
-            // Normalize all weights in each individual head
             for (const head of attentionHeads) {
                 const allWeightsForHead = head.flatMap(tokenFrames => tokenFrames);
-                const meanOfAllWeights = meanOfVector(allWeightsForHead);
-                const stdDeviationOfAllWeights = stdDeviationOfVector(allWeightsForHead) + 1e-10;
+                const meanOfAllWeightsForHead = meanOfVector(allWeightsForHead);
+                const stdDeviationOfAllWeightsForHead = stdDeviationOfVector(allWeightsForHead, 'population', meanOfAllWeightsForHead) + 1e-10;
+                const stdDeviationReciprocal = 1.0 / (stdDeviationOfAllWeightsForHead + 1e-10);
                 for (const tokenFrames of head) {
                     for (let frameIndex = 0; frameIndex < tokenFrames.length; frameIndex++) {
-                        tokenFrames[frameIndex] = (tokenFrames[frameIndex] - meanOfAllWeights) / stdDeviationOfAllWeights;
+                        tokenFrames[frameIndex] = (tokenFrames[frameIndex] - meanOfAllWeightsForHead) * stdDeviationReciprocal;
                     }
                 }
             }
         }
+        // Apply median filter to each token's frames, if enabled
         if (applyMedianFilter) {
-            // Apply median filter to each token's frames
             for (const head of attentionHeads) {
                 for (let tokenIndex = 0; tokenIndex < tokenCount; tokenIndex++) {
-                    head[tokenIndex] = medianFilter(head[tokenIndex], medianFilterWidth);
+                    head[tokenIndex] = medianOf5Filter(head[tokenIndex]);
                 }
             }
         }
@@ -869,8 +919,8 @@ export class Whisper {
                 frameMeansForToken[tokenIndex][frameIndex] = frameMean;
             }
         }
-        if (fixateTimestampTokens) {
-            // Fixate timestamp tokens to the original ones detected
+        // Anchor timestamp tokens timestamps to their original values, if enabled
+        if (anchorTimestampTokens) {
             const timestampTokensStart = this.tokenConfig.timestampTokensStart;
             for (let tokenIndex = 0; tokenIndex < tokens.length; tokenIndex++) {
                 const token = tokens[tokenIndex];
@@ -882,14 +932,86 @@ export class Whisper {
             }
         }
         // Perform DTW
-        const tokenIndexes = [...Array(tokenCount).keys()];
-        const frameIndexes = [...Array(segmentFrameCount).keys()];
+        const tokenIndexes = getIntegerRange(0, tokenCount);
+        const frameIndexes = getIntegerRange(0, segmentFrameCount);
         let { path } = alignDTWWindowed(tokenIndexes, frameIndexes, (tokenIndex, frameIndex) => {
             return -frameMeansForToken[tokenIndex][frameIndex];
         }, segmentFrameCount);
         path = path.map(entry => ({ source: entry.source, dest: segmentStartFrame + entry.dest }));
         return path;
     }
+    async initializeIfNeeded() {
+        await this.initializeTokenizerIfNeeded();
+        await this.initializeEncoderSessionIfNeeded();
+        await this.initializeDecoderSessionIfNeeded();
+    }
+    async initializeTokenizerIfNeeded() {
+        if (this.tiktoken) {
+            return;
+        }
+        const logger = new Logger();
+        await logger.startAsync('Load tokenizer data');
+        const tiktokenModulePackagePath = await loadPackage('whisper-tiktoken-data');
+        const tiktokenDataFilePath = path.join(tiktokenModulePackagePath, this.isMultiligualModel ? 'multilingual.tiktoken' : 'gpt2.tiktoken');
+        let tiktokenData = await readFile(tiktokenDataFilePath, { encoding: 'utf8' });
+        const tokenConfig = this.tokenConfig;
+        const metadataTokens = {
+            [tokenConfig.endOfTextToken]: '[EndOfText]',
+            [tokenConfig.startOfTextToken]: '[StartOfText]',
+            [tokenConfig.translateTaskToken]: '[TranslateTask]',
+            [tokenConfig.transcribeTaskToken]: '[TranscribeTask]',
+            [tokenConfig.startOfPromptToken]: '[StartOfPrompt]',
+            [tokenConfig.nonSpeechToken]: '[NonSpeech]',
+            [tokenConfig.noTimestampsToken]: '[NoTimestamps]',
+        };
+        if (this.isMultiligualModel) {
+            metadataTokens[50256] = '[Unused_50256]';
+            metadataTokens[50360] = '[Unused_50360]';
+        }
+        const languageTokenCount = tokenConfig.languageTokensEnd - tokenConfig.languageTokensStart;
+        for (let i = 0; i < languageTokenCount; i++) {
+            const tokenIndex = this.tokenConfig.languageTokensStart + i;
+            metadataTokens[tokenIndex] = `[Language_${i}]`;
+        }
+        const timestampTokensCount = 1501;
+        for (let i = 0; i < timestampTokensCount; i++) {
+            const tokenIndex = this.tokenConfig.timestampTokensStart + i;
+            const tokenTime = this.timestampTokenToSeconds(tokenIndex);
+            metadataTokens[tokenIndex] = `[Timestamp_${tokenTime.toFixed(2)}]`;
+        }
+        const inverseMetadataTokensLookup = {};
+        for (const [key, value] of Object.entries(metadataTokens)) {
+            inverseMetadataTokensLookup[value] = parseInt(key);
+        }
+        const patternString = `'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)|\\s+`;
+        const { Tiktoken } = await import('tiktoken/lite');
+        this.tiktoken = new Tiktoken(tiktokenData, inverseMetadataTokensLookup, patternString);
+        logger.end();
+    }
+    async initializeEncoderSessionIfNeeded() {
+        if (this.audioEncoder) {
+            return;
+        }
+        const logger = new Logger();
+        await logger.startAsync(`Create encoder inference session for model '${this.modelName}'`);
+        const encoderFilePath = path.join(this.modelDir, 'encoder.onnx');
+        const Onnx = await import('onnxruntime-node');
+        const onnxSessionOptions = getOnnxSessionOptions({ executionProviders: this.encoderExecutionProviders });
+        this.audioEncoder = await Onnx.InferenceSession.create(encoderFilePath, onnxSessionOptions);
+        logger.end();
+    }
+    async initializeDecoderSessionIfNeeded() {
+        if (this.textDecoder) {
+            return;
+        }
+        const logger = new Logger();
+        await logger.startAsync(`Create decoder inference session for model '${this.modelName}'`);
+        const decoderFilePath = path.join(this.modelDir, 'decoder.onnx');
+        const Onnx = await import('onnxruntime-node');
+        const onnxSessionOptions = getOnnxSessionOptions({ executionProviders: this.decoderExecutionProviders });
+        this.textDecoder = await Onnx.InferenceSession.create(decoderFilePath, onnxSessionOptions);
+        logger.end();
+    }
     getKvDimensions(groupCount, length) {
         const modelName = this.modelName;
         if (modelName == 'tiny' || modelName == 'tiny.en') {
@@ -1010,7 +1132,7 @@ export class Whisper {
     }
     getSuppressedTextTokens() {
         const allowedPunctuationMarks = this.getAllowedPunctuationMarks();
-        const nonWordTokensData = this.getNonWordTokenData();
+        const nonWordTokensData = this.getWordTokenData().nonWordTokenData;
         const suppressedTextTokens = nonWordTokensData
             .filter(entry => !allowedPunctuationMarks.includes(entry.text))
             .map(entry => entry.id);
@@ -1039,22 +1161,27 @@ export class Whisper {
         }
         return allowedPunctuation;
     }
-    getNonWordTokenData() {
+    getWordTokenData() {
+        const wordTokenData = [];
         const nonWordTokenData = [];
-        const invalidUTF8Char = String.fromCharCode(65533);
         for (let i = 0; i < this.tokenConfig.endOfTextToken; i++) {
             const tokenText = this.tokenToText(i, false);
-            const tokenTextWithoutWhitespace = tokenText.replaceAll(/\s/g, '');
-            const isNonWordToken = /^[\p{Punctuation}\p{Symbol}]+$/u.test(tokenTextWithoutWhitespace);
-            const containsInvalidUTF8 = getUTF32Chars(tokenTextWithoutWhitespace).utf32chars.includes(invalidUTF8Char);
-            if (isNonWordToken && !containsInvalidUTF8) {
+            const isNonWordToken = /^[\s\p{Punctuation}\p{Symbol}]+$/u.test(tokenText);
+            const containsInvalidUTF8 = containsInvalidCodepoint(tokenText);
+            if (isNonWordToken && (this.isEnglishOnlyModel || !containsInvalidUTF8)) {
                 nonWordTokenData.push({
                     id: i,
                     text: tokenText,
                 });
             }
+            else {
+                wordTokenData.push({
+                    id: i,
+                    text: tokenText,
+                });
+            }
         }
-        return nonWordTokenData;
+        return { wordTokenData, nonWordTokenData };
     }
     getTokensData(tokens) {
         const tokensData = [];
@@ -1067,88 +1194,6 @@ export class Whisper {
         return tokensData;
     }
 }
-const filterbanks = [
-    /* 0 */ { startIndex: 1, weights: [0.02486259490251541,] },
-    /* 1 */ { startIndex: 1, weights: [0.001990821911022067, 0.022871771827340126,] },
-    /* 2 */ { startIndex: 2, weights: [0.003981643822044134, 0.02088095061480999,] },
-    /* 3 */ { startIndex: 3, weights: [0.0059724655002355576, 0.018890129402279854,] },
-    /* 4 */ { startIndex: 4, weights: [0.007963287644088268, 0.01689930632710457,] },
-    /* 5 */ { startIndex: 5, weights: [0.009954108856618404, 0.014908484183251858,] },
-    /* 6 */ { startIndex: 6, weights: [0.011944931000471115, 0.012917662039399147,] },
-    /* 7 */ { startIndex: 7, weights: [0.013935752213001251, 0.010926840826869011,] },
-    /* 8 */ { startIndex: 8, weights: [0.015926575288176537, 0.0089360186830163,] },
-    /* 9 */ { startIndex: 9, weights: [0.017917396500706673, 0.006945197004824877,] },
-    /* 10 */ { startIndex: 10, weights: [0.01990821771323681, 0.004954374860972166,] },
-    /* 11 */ { startIndex: 11, weights: [0.021899040788412094, 0.0029635531827807426,] },
-    /* 12 */ { startIndex: 12, weights: [0.02388986200094223, 0.0009727313299663365,] },
-    /* 13 */ { startIndex: 13, weights: [0.025880683213472366,] },
-    /* 14 */ { startIndex: 14, weights: [0.025835324078798294,] },
-    /* 15 */ { startIndex: 14, weights: [0.0010180906392633915, 0.023844502866268158,] },
-    /* 16 */ { startIndex: 15, weights: [0.003008912317454815, 0.021853681653738022,] },
-    /* 17 */ { startIndex: 16, weights: [0.004999734461307526, 0.019862858578562737,] },
-    /* 18 */ { startIndex: 17, weights: [0.006990555673837662, 0.0178720373660326,] },
-    /* 19 */ { startIndex: 18, weights: [0.008981377817690372, 0.015881216153502464,] },
-    /* 20 */ { startIndex: 19, weights: [0.010972199961543083, 0.013890394009649754,] },
-    /* 21 */ { startIndex: 20, weights: [0.01296302117407322, 0.011899571865797043,] },
-    /* 22 */ { startIndex: 21, weights: [0.01495384331792593, 0.009908749721944332,] },
-    /* 23 */ { startIndex: 22, weights: [0.01694466546177864, 0.007917927578091621,] },
-    /* 24 */ { startIndex: 23, weights: [0.018935488536953926, 0.005927106365561485,] },
-    /* 25 */ { startIndex: 24, weights: [0.020874010398983955, 0.004040425643324852,] },
-    /* 26 */ { startIndex: 25, weights: [0.022114217281341553, 0.0033186059445142746,] },
-    /* 27 */ { startIndex: 26, weights: [0.02173672430217266, 0.0036109676584601402,] },
-    /* 28 */ { startIndex: 27, weights: [0.020497702062129974, 0.004762193653732538,] },
-    /* 29 */ { startIndex: 28, weights: [0.018486659973859787, 0.006592618301510811,] },
-    /* 30 */ { startIndex: 29, weights: [0.01585603691637516, 0.00896277092397213,] },
-    /* 31 */ { startIndex: 30, weights: [0.012738768011331558, 0.011751330457627773,] },
-    /* 32 */ { startIndex: 31, weights: [0.009250369854271412, 0.014853144995868206,] },
-    /* 33 */ { startIndex: 32, weights: [0.005490840878337622, 0.018177473917603493, 0.0028155462350696325,] },
-    /* 34 */ { startIndex: 33, weights: [0.0015463664894923568, 0.01632951945066452, 0.007420188747346401,] },
-    /* 35 */ { startIndex: 35, weights: [0.011181050911545753, 0.012018864043056965,] },
-    /* 36 */ { startIndex: 36, weights: [0.006065350491553545, 0.016561277210712433, 0.004360878840088844,] },
-    /* 37 */ { startIndex: 37, weights: [0.0010297985281795263, 0.012770536355674267, 0.009707189165055752,] },
-    /* 38 */ { startIndex: 39, weights: [0.006986402906477451, 0.01485429983586073, 0.004391219466924667,] },
-    /* 39 */ { startIndex: 40, weights: [0.001418047584593296, 0.011486922390758991, 0.010089744813740253, 0.00040022286702878773,] },
-    /* 40 */ { startIndex: 42, weights: [0.005411104764789343, 0.014735566452145576, 0.006518189795315266,] },
-    /* 41 */ { startIndex: 44, weights: [0.00827841367572546, 0.012277561239898205, 0.00396781275048852,] },
-    /* 42 */ { startIndex: 45, weights: [0.002187808509916067, 0.010184479877352715, 0.00998187530785799, 0.0022864851634949446,] },
-    /* 43 */ { startIndex: 47, weights: [0.00386943481862545, 0.011274894699454308, 0.008466221392154694, 0.0013397691072896123,] },
-    /* 44 */ { startIndex: 49, weights: [0.004820294212549925, 0.011678251437842846, 0.007608682848513126, 0.0010091039584949613,] },
-    /* 45 */ { startIndex: 51, weights: [0.005156961735337973, 0.011507894843816757, 0.007301822770386934, 0.0011901655234396458,] },
-    /* 46 */ { startIndex: 53, weights: [0.004982104524970055, 0.010863498784601688, 0.007451189681887627, 0.001791381393559277,] },
-    /* 47 */ { startIndex: 55, weights: [0.004385921638458967, 0.009832492098212242, 0.007973956875503063, 0.002732589840888977,] },
-    /* 48 */ { startIndex: 57, weights: [0.0034474546555429697, 0.008491347543895245, 0.008797688409686089, 0.00394382793456316,] },
-    /* 49 */ { startIndex: 59, weights: [0.0022357646375894547, 0.0069067515432834625, 0.009859241545200348, 0.005364237818866968, 0.0008692338014952838,] },
-    /* 50 */ { startIndex: 61, weights: [0.0008110002381727099, 0.005136650986969471, 0.00946230161935091, 0.0069410777650773525, 0.0027783995028585196,] },
-    /* 51 */ { startIndex: 64, weights: [0.003231203882023692, 0.007237049750983715, 0.00862883497029543, 0.004773912951350212, 0.0009189908159896731,] },
-    /* 52 */ { startIndex: 66, weights: [0.001233637798577547, 0.0049433219246566296, 0.008653006516397, 0.006818502210080624, 0.003248583758249879,] },
-    /* 53 */ { startIndex: 69, weights: [0.0026164355222135782, 0.006051854696124792, 0.008880467154085636, 0.005574479699134827, 0.002268492942675948,] },
-    /* 54 */ { startIndex: 71, weights: [0.0002863667905330658, 0.003467798000201583, 0.0066492292098701, 0.00787146482616663, 0.004809896927326918, 0.0017483289120718837,] },
-    /* 55 */ { startIndex: 74, weights: [0.0009245910914614797, 0.0038708120118826628, 0.00681703258305788, 0.007283343467861414, 0.004448124207556248, 0.0016129047144204378,] },
-    /* 56 */ { startIndex: 77, weights: [0.0011703289346769452, 0.003898728871718049, 0.006627128925174475, 0.0070473202504217625, 0.004421714693307877, 0.0017961094854399562,] },
-    /* 57 */ { startIndex: 80, weights: [0.0010892992140725255, 0.003615982597693801, 0.006142666097730398, 0.007102936040610075, 0.004671447444707155, 0.002239959081634879,] },
-    /* 58 */ { startIndex: 83, weights: [0.0007392280967906117, 0.0030791081953793764, 0.005418988410383463, 0.007397185545414686, 0.005145462695509195, 0.002893739379942417, 0.0006420162972062826,] },
-    /* 59 */ { startIndex: 86, weights: [0.00017068670422304422, 0.0023375742603093386, 0.004504461772739887, 0.0066713495180010796, 0.005798479542136192, 0.003713231300935149, 0.0016279831761494279,] },
-    /* 60 */ { startIndex: 90, weights: [0.0014345343224704266, 0.0034412189852446318, 0.005447904113680124, 0.006591092795133591, 0.004660011734813452, 0.002728930441662669, 0.0007978491485118866,] },
-    /* 61 */ { startIndex: 93, weights: [0.0004075043834745884, 0.002265830524265766, 0.004124156199395657, 0.005982482805848122, 0.005700822453945875, 0.003912510350346565, 0.0021241982467472553, 0.0003358862304594368,] },
-    /* 62 */ { startIndex: 97, weights: [0.0010099108330905437, 0.002730846870690584, 0.004451782442629337, 0.006172718480229378, 0.005150905344635248, 0.0034948070533573627, 0.0018387088784947991, 0.0001826105872169137,] },
-    /* 63 */ { startIndex: 101, weights: [0.0012943691108375788, 0.002888072282075882, 0.004481775686144829, 0.006075479090213776, 0.0048866597935557365, 0.003353001084178686, 0.0018193417927250266, 0.00028568264679051936,] },
-    /* 64 */ { startIndex: 105, weights: [0.0013131388695910573, 0.0027890161145478487, 0.004264893010258675, 0.0057407706044614315, 0.004859979264438152, 0.0034397069830447435, 0.0020194342359900475, 0.0005991620710119605,] },
-    /* 65 */ { startIndex: 109, weights: [0.0011121684219688177, 0.002478930866345763, 0.0038456933107227087, 0.0052124555222690105, 0.005028639920055866, 0.0037133716978132725, 0.002398103242740035, 0.0010828346712514758,] },
-    /* 66 */ { startIndex: 113, weights: [0.0007317548734135926, 0.0019974694587290287, 0.003263183869421482, 0.004528898745775223, 0.005355686880648136, 0.004137659445405006, 0.0029196315445005894, 0.0017016039928421378, 0.0004835762665607035,] },
-    /* 67 */ { startIndex: 117, weights: [0.00020713974663522094, 0.0013792773243039846, 0.0025514145381748676, 0.003723552217707038, 0.004895689897239208, 0.004680895246565342, 0.0035529187880456448, 0.0024249425623565912, 0.0012969663366675377, 0.00016899015463422984,] },
-    /* 68 */ { startIndex: 122, weights: [0.0006545265205204487, 0.0017400053329765797, 0.0028254841454327106, 0.003910962492227554, 0.004996441304683685, 0.0042709787376224995, 0.003226396394893527, 0.002181813819333911, 0.0011372314766049385, 9.264905384043232e-05,] },
-    /* 69 */ { startIndex: 127, weights: [0.000854626705404371, 0.001859853626228869, 0.002865080488845706, 0.003870307235047221, 0.00487553421407938, 0.00408313749358058, 0.003115783678367734, 0.0021484296303242445, 0.001181075582280755, 0.0002137213887181133,] },
-    /* 70 */ { startIndex: 132, weights: [0.0008483415003865957, 0.0017792496364563704, 0.0027101580053567886, 0.0036410661414265633, 0.004571974277496338, 0.004079728852957487, 0.003183893393725157, 0.002288057701662183, 0.0013922222424298525, 0.0004963868414051831,] },
-    /* 71 */ { startIndex: 137, weights: [0.0006716204807162285, 0.0015337044605985284, 0.002395788673311472, 0.0032578727696090937, 0.004119956865906715, 0.004227725323289633, 0.0033981208689510822, 0.0025685166474431753, 0.0017389123095199466, 0.0009093079133890569, 7.970355363795534e-05,] },
-    /* 72 */ { startIndex: 142, weights: [0.0003559796023182571, 0.0011543278815224767, 0.0019526762189343572, 0.002751024439930916, 0.0035493727773427963, 0.004347721114754677, 0.0037299629766494036, 0.002961693098768592, 0.00219342322088778, 0.0014251532265916467, 0.0006568834069184959,] },
-    /* 73 */ { startIndex: 148, weights: [0.0006682946113869548, 0.0014076193328946829, 0.0021469437051564455, 0.002886268775910139, 0.0036255933810025454, 0.004154576454311609, 0.0034431067761033773, 0.0027316368650645018, 0.0020201667211949825, 0.0013086966937407851, 0.0005972267827019095,] },
-    /* 74 */ { startIndex: 153, weights: [9.926508937496692e-05, 0.0007839298341423273, 0.001468594535253942, 0.0021532592363655567, 0.0028379240538924932, 0.0035225888714194298, 0.0039915177039802074, 0.0033326479606330395, 0.002673778682947159, 0.002014909405261278, 0.0013560398947447538, 0.0006971705006435513, 3.8301113818306476e-05,] },
-    /* 75 */ { startIndex: 159, weights: [0.00010181095422012731, 0.0007358568836934865, 0.0013699028640985489, 0.0020039486698806286, 0.002637994708493352, 0.0032720407471060753, 0.003906086552888155, 0.0033682563807815313, 0.0027580985333770514, 0.002147940918803215, 0.0015377833042293787, 0.0009276255150325596, 0.000317467754939571,] },
-    /* 76 */ { startIndex: 166, weights: [0.0005530364578589797, 0.0011402058880776167, 0.0017273754347115755, 0.0023145449813455343, 0.002901714527979493, 0.003488884074613452, 0.003523340215906501, 0.002958292607218027, 0.002393245231360197, 0.0018281979719176888, 0.001263150479644537, 0.0006981031037867069, 0.0001330557424807921,] },
-    /* 77 */ { startIndex: 172, weights: [0.0002608386566862464, 0.0008045974536798894, 0.0013483562506735325, 0.0018921148730441928, 0.0024358737282454967, 0.002979632467031479, 0.003523391205817461, 0.003251380519941449, 0.0027281083166599274, 0.002204835880547762, 0.001681563793681562, 0.001158291706815362, 0.0006350195035338402, 0.00011174729297636077,] },
-    /* 78 */ { startIndex: 179, weights: [0.0003849811910185963, 0.0008885387214832008, 0.001392096164636314, 0.0018956535495817661, 0.00239921105094254, 0.002902768552303314, 0.0034063260536640882, 0.003132763085886836, 0.0026481777895241976, 0.0021635922603309155, 0.0016790067311376333, 0.0011944210855290294, 0.0007098356145434082, 0.00022525011445395648,] },
-    /* 79 */ { startIndex: 186, weights: [0.000366741674952209, 0.0008330700220540166, 0.0012993983691558242, 0.0017657268326729536, 0.0022320549469441175, 0.002698383294045925, 0.0031647118739783764, 0.003141313325613737, 0.002692554146051407, 0.0022437951993197203, 0.00179503601975739, 0.0013462770730257034, 0.000897518009878695, 0.0004487590049393475,] },
-];
 export async function loadPackagesAndGetPaths(modelName, languageCode) {
     if (modelName) {
         modelName = normalizeWhisperModelName(modelName, languageCode);
@@ -1314,6 +1359,88 @@ const alignmentHeadsIndexes = {
     'large-v3': [212, 277, 331, 332, 333, 355, 356, 364, 371, 379, 391, 422, 423, 443, 449, 452, 465, 467, 473, 505, 521, 532, 555], // Temporary (may not be correct)
     'large': [212, 277, 331, 332, 333, 355, 356, 364, 371, 379, 391, 422, 423, 443, 449, 452, 465, 467, 473, 505, 521, 532, 555],
 };
+const filterbanks = [
+    /* 0 */ { startIndex: 1, weights: [0.02486259490251541,] },
+    /* 1 */ { startIndex: 1, weights: [0.001990821911022067, 0.022871771827340126,] },
+    /* 2 */ { startIndex: 2, weights: [0.003981643822044134, 0.02088095061480999,] },
+    /* 3 */ { startIndex: 3, weights: [0.0059724655002355576, 0.018890129402279854,] },
+    /* 4 */ { startIndex: 4, weights: [0.007963287644088268, 0.01689930632710457,] },
+    /* 5 */ { startIndex: 5, weights: [0.009954108856618404, 0.014908484183251858,] },
+    /* 6 */ { startIndex: 6, weights: [0.011944931000471115, 0.012917662039399147,] },
+    /* 7 */ { startIndex: 7, weights: [0.013935752213001251, 0.010926840826869011,] },
+    /* 8 */ { startIndex: 8, weights: [0.015926575288176537, 0.0089360186830163,] },
+    /* 9 */ { startIndex: 9, weights: [0.017917396500706673, 0.006945197004824877,] },
+    /* 10 */ { startIndex: 10, weights: [0.01990821771323681, 0.004954374860972166,] },
+    /* 11 */ { startIndex: 11, weights: [0.021899040788412094, 0.0029635531827807426,] },
+    /* 12 */ { startIndex: 12, weights: [0.02388986200094223, 0.0009727313299663365,] },
+    /* 13 */ { startIndex: 13, weights: [0.025880683213472366,] },
+    /* 14 */ { startIndex: 14, weights: [0.025835324078798294,] },
+    /* 15 */ { startIndex: 14, weights: [0.0010180906392633915, 0.023844502866268158,] },
+    /* 16 */ { startIndex: 15, weights: [0.003008912317454815, 0.021853681653738022,] },
+    /* 17 */ { startIndex: 16, weights: [0.004999734461307526, 0.019862858578562737,] },
+    /* 18 */ { startIndex: 17, weights: [0.006990555673837662, 0.0178720373660326,] },
+    /* 19 */ { startIndex: 18, weights: [0.008981377817690372, 0.015881216153502464,] },
+    /* 20 */ { startIndex: 19, weights: [0.010972199961543083, 0.013890394009649754,] },
+    /* 21 */ { startIndex: 20, weights: [0.01296302117407322, 0.011899571865797043,] },
+    /* 22 */ { startIndex: 21, weights: [0.01495384331792593, 0.009908749721944332,] },
+    /* 23 */ { startIndex: 22, weights: [0.01694466546177864, 0.007917927578091621,] },
+    /* 24 */ { startIndex: 23, weights: [0.018935488536953926, 0.005927106365561485,] },
+    /* 25 */ { startIndex: 24, weights: [0.020874010398983955, 0.004040425643324852,] },
+    /* 26 */ { startIndex: 25, weights: [0.022114217281341553, 0.0033186059445142746,] },
+    /* 27 */ { startIndex: 26, weights: [0.02173672430217266, 0.0036109676584601402,] },
+    /* 28 */ { startIndex: 27, weights: [0.020497702062129974, 0.004762193653732538,] },
+    /* 29 */ { startIndex: 28, weights: [0.018486659973859787, 0.006592618301510811,] },
+    /* 30 */ { startIndex: 29, weights: [0.01585603691637516, 0.00896277092397213,] },
+    /* 31 */ { startIndex: 30, weights: [0.012738768011331558, 0.011751330457627773,] },
+    /* 32 */ { startIndex: 31, weights: [0.009250369854271412, 0.014853144995868206,] },
+    /* 33 */ { startIndex: 32, weights: [0.005490840878337622, 0.018177473917603493, 0.0028155462350696325,] },
+    /* 34 */ { startIndex: 33, weights: [0.0015463664894923568, 0.01632951945066452, 0.007420188747346401,] },
+    /* 35 */ { startIndex: 35, weights: [0.011181050911545753, 0.012018864043056965,] },
+    /* 36 */ { startIndex: 36, weights: [0.006065350491553545, 0.016561277210712433, 0.004360878840088844,] },
+    /* 37 */ { startIndex: 37, weights: [0.0010297985281795263, 0.012770536355674267, 0.009707189165055752,] },
+    /* 38 */ { startIndex: 39, weights: [0.006986402906477451, 0.01485429983586073, 0.004391219466924667,] },
+    /* 39 */ { startIndex: 40, weights: [0.001418047584593296, 0.011486922390758991, 0.010089744813740253, 0.00040022286702878773,] },
+    /* 40 */ { startIndex: 42, weights: [0.005411104764789343, 0.014735566452145576, 0.006518189795315266,] },
+    /* 41 */ { startIndex: 44, weights: [0.00827841367572546, 0.012277561239898205, 0.00396781275048852,] },
+    /* 42 */ { startIndex: 45, weights: [0.002187808509916067, 0.010184479877352715, 0.00998187530785799, 0.0022864851634949446,] },
+    /* 43 */ { startIndex: 47, weights: [0.00386943481862545, 0.011274894699454308, 0.008466221392154694, 0.0013397691072896123,] },
+    /* 44 */ { startIndex: 49, weights: [0.004820294212549925, 0.011678251437842846, 0.007608682848513126, 0.0010091039584949613,] },
+    /* 45 */ { startIndex: 51, weights: [0.005156961735337973, 0.011507894843816757, 0.007301822770386934, 0.0011901655234396458,] },
+    /* 46 */ { startIndex: 53, weights: [0.004982104524970055, 0.010863498784601688, 0.007451189681887627, 0.001791381393559277,] },
+    /* 47 */ { startIndex: 55, weights: [0.004385921638458967, 0.009832492098212242, 0.007973956875503063, 0.002732589840888977,] },
+    /* 48 */ { startIndex: 57, weights: [0.0034474546555429697, 0.008491347543895245, 0.008797688409686089, 0.00394382793456316,] },
+    /* 49 */ { startIndex: 59, weights: [0.0022357646375894547, 0.0069067515432834625, 0.009859241545200348, 0.005364237818866968, 0.0008692338014952838,] },
+    /* 50 */ { startIndex: 61, weights: [0.0008110002381727099, 0.005136650986969471, 0.00946230161935091, 0.0069410777650773525, 0.0027783995028585196,] },
+    /* 51 */ { startIndex: 64, weights: [0.003231203882023692, 0.007237049750983715, 0.00862883497029543, 0.004773912951350212, 0.0009189908159896731,] },
+    /* 52 */ { startIndex: 66, weights: [0.001233637798577547, 0.0049433219246566296, 0.008653006516397, 0.006818502210080624, 0.003248583758249879,] },
+    /* 53 */ { startIndex: 69, weights: [0.0026164355222135782, 0.006051854696124792, 0.008880467154085636, 0.005574479699134827, 0.002268492942675948,] },
+    /* 54 */ { startIndex: 71, weights: [0.0002863667905330658, 0.003467798000201583, 0.0066492292098701, 0.00787146482616663, 0.004809896927326918, 0.0017483289120718837,] },
+    /* 55 */ { startIndex: 74, weights: [0.0009245910914614797, 0.0038708120118826628, 0.00681703258305788, 0.007283343467861414, 0.004448124207556248, 0.0016129047144204378,] },
+    /* 56 */ { startIndex: 77, weights: [0.0011703289346769452, 0.003898728871718049, 0.006627128925174475, 0.0070473202504217625, 0.004421714693307877, 0.0017961094854399562,] },
+    /* 57 */ { startIndex: 80, weights: [0.0010892992140725255, 0.003615982597693801, 0.006142666097730398, 0.007102936040610075, 0.004671447444707155, 0.002239959081634879,] },
+    /* 58 */ { startIndex: 83, weights: [0.0007392280967906117, 0.0030791081953793764, 0.005418988410383463, 0.007397185545414686, 0.005145462695509195, 0.002893739379942417, 0.0006420162972062826,] },
+    /* 59 */ { startIndex: 86, weights: [0.00017068670422304422, 0.0023375742603093386, 0.004504461772739887, 0.0066713495180010796, 0.005798479542136192, 0.003713231300935149, 0.0016279831761494279,] },
+    /* 60 */ { startIndex: 90, weights: [0.0014345343224704266, 0.0034412189852446318, 0.005447904113680124, 0.006591092795133591, 0.004660011734813452, 0.002728930441662669, 0.0007978491485118866,] },
+    /* 61 */ { startIndex: 93, weights: [0.0004075043834745884, 0.002265830524265766, 0.004124156199395657, 0.005982482805848122, 0.005700822453945875, 0.003912510350346565, 0.0021241982467472553, 0.0003358862304594368,] },
+    /* 62 */ { startIndex: 97, weights: [0.0010099108330905437, 0.002730846870690584, 0.004451782442629337, 0.006172718480229378, 0.005150905344635248, 0.0034948070533573627, 0.0018387088784947991, 0.0001826105872169137,] },
+    /* 63 */ { startIndex: 101, weights: [0.0012943691108375788, 0.002888072282075882, 0.004481775686144829, 0.006075479090213776, 0.0048866597935557365, 0.003353001084178686, 0.0018193417927250266, 0.00028568264679051936,] },
+    /* 64 */ { startIndex: 105, weights: [0.0013131388695910573, 0.0027890161145478487, 0.004264893010258675, 0.0057407706044614315, 0.004859979264438152, 0.0034397069830447435, 0.0020194342359900475, 0.0005991620710119605,] },
+    /* 65 */ { startIndex: 109, weights: [0.0011121684219688177, 0.002478930866345763, 0.0038456933107227087, 0.0052124555222690105, 0.005028639920055866, 0.0037133716978132725, 0.002398103242740035, 0.0010828346712514758,] },
+    /* 66 */ { startIndex: 113, weights: [0.0007317548734135926, 0.0019974694587290287, 0.003263183869421482, 0.004528898745775223, 0.005355686880648136, 0.004137659445405006, 0.0029196315445005894, 0.0017016039928421378, 0.0004835762665607035,] },
+    /* 67 */ { startIndex: 117, weights: [0.00020713974663522094, 0.0013792773243039846, 0.0025514145381748676, 0.003723552217707038, 0.004895689897239208, 0.004680895246565342, 0.0035529187880456448, 0.0024249425623565912, 0.0012969663366675377, 0.00016899015463422984,] },
+    /* 68 */ { startIndex: 122, weights: [0.0006545265205204487, 0.0017400053329765797, 0.0028254841454327106, 0.003910962492227554, 0.004996441304683685, 0.0042709787376224995, 0.003226396394893527, 0.002181813819333911, 0.0011372314766049385, 9.264905384043232e-05,] },
+    /* 69 */ { startIndex: 127, weights: [0.000854626705404371, 0.001859853626228869, 0.002865080488845706, 0.003870307235047221, 0.00487553421407938, 0.00408313749358058, 0.003115783678367734, 0.0021484296303242445, 0.001181075582280755, 0.0002137213887181133,] },
+    /* 70 */ { startIndex: 132, weights: [0.0008483415003865957, 0.0017792496364563704, 0.0027101580053567886, 0.0036410661414265633, 0.004571974277496338, 0.004079728852957487, 0.003183893393725157, 0.002288057701662183, 0.0013922222424298525, 0.0004963868414051831,] },
+    /* 71 */ { startIndex: 137, weights: [0.0006716204807162285, 0.0015337044605985284, 0.002395788673311472, 0.0032578727696090937, 0.004119956865906715, 0.004227725323289633, 0.0033981208689510822, 0.0025685166474431753, 0.0017389123095199466, 0.0009093079133890569, 7.970355363795534e-05,] },
+    /* 72 */ { startIndex: 142, weights: [0.0003559796023182571, 0.0011543278815224767, 0.0019526762189343572, 0.002751024439930916, 0.0035493727773427963, 0.004347721114754677, 0.0037299629766494036, 0.002961693098768592, 0.00219342322088778, 0.0014251532265916467, 0.0006568834069184959,] },
+    /* 73 */ { startIndex: 148, weights: [0.0006682946113869548, 0.0014076193328946829, 0.0021469437051564455, 0.002886268775910139, 0.0036255933810025454, 0.004154576454311609, 0.0034431067761033773, 0.0027316368650645018, 0.0020201667211949825, 0.0013086966937407851, 0.0005972267827019095,] },
+    /* 74 */ { startIndex: 153, weights: [9.926508937496692e-05, 0.0007839298341423273, 0.001468594535253942, 0.0021532592363655567, 0.0028379240538924932, 0.0035225888714194298, 0.0039915177039802074, 0.0033326479606330395, 0.002673778682947159, 0.002014909405261278, 0.0013560398947447538, 0.0006971705006435513, 3.8301113818306476e-05,] },
+    /* 75 */ { startIndex: 159, weights: [0.00010181095422012731, 0.0007358568836934865, 0.0013699028640985489, 0.0020039486698806286, 0.002637994708493352, 0.0032720407471060753, 0.003906086552888155, 0.0033682563807815313, 0.0027580985333770514, 0.002147940918803215, 0.0015377833042293787, 0.0009276255150325596, 0.000317467754939571,] },
+    /* 76 */ { startIndex: 166, weights: [0.0005530364578589797, 0.0011402058880776167, 0.0017273754347115755, 0.0023145449813455343, 0.002901714527979493, 0.003488884074613452, 0.003523340215906501, 0.002958292607218027, 0.002393245231360197, 0.0018281979719176888, 0.001263150479644537, 0.0006981031037867069, 0.0001330557424807921,] },
+    /* 77 */ { startIndex: 172, weights: [0.0002608386566862464, 0.0008045974536798894, 0.0013483562506735325, 0.0018921148730441928, 0.0024358737282454967, 0.002979632467031479, 0.003523391205817461, 0.003251380519941449, 0.0027281083166599274, 0.002204835880547762, 0.001681563793681562, 0.001158291706815362, 0.0006350195035338402, 0.00011174729297636077,] },
+    /* 78 */ { startIndex: 179, weights: [0.0003849811910185963, 0.0008885387214832008, 0.001392096164636314, 0.0018956535495817661, 0.00239921105094254, 0.002902768552303314, 0.0034063260536640882, 0.003132763085886836, 0.0026481777895241976, 0.0021635922603309155, 0.0016790067311376333, 0.0011944210855290294, 0.0007098356145434082, 0.00022525011445395648,] },
+    /* 79 */ { startIndex: 186, weights: [0.000366741674952209, 0.0008330700220540166, 0.0012993983691558242, 0.0017657268326729536, 0.0022320549469441175, 0.002698383294045925, 0.0031647118739783764, 0.003141313325613737, 0.002692554146051407, 0.0022437951993197203, 0.00179503601975739, 0.0013462770730257034, 0.000897518009878695, 0.0004487590049393475,] },
+];
 export const defaultWhisperOptions = {
     model: undefined,
     temperature: 0.1,
@@ -1324,6 +1451,28 @@ export const defaultWhisperOptions = {
     maxTokensPerPart: 250,
     suppressRepetition: true,
     decodeTimestampTokens: true,
+    endTokenThreshold: 0.9,
+    includeEndTokenInCandidates: true,
+    encoderProvider: undefined,
+    decoderProvider: undefined,
     seed: undefined,
 };
+export const defaultWhisperAlignmentOptions = {
+    model: undefined,
+    endTokenThreshold: 0.9,
+    encoderProvider: undefined,
+    decoderProvider: undefined
+};
+export const defaultWhisperLanguageDetectionOptions = {
+    model: undefined,
+    temperature: 1.0,
+    encoderProvider: undefined,
+    decoderProvider: undefined,
+};
+export const defaultWhisperVADOptions = {
+    model: undefined,
+    temperature: 1.0,
+    encoderProvider: undefined,
+    decoderProvider: undefined,
+};
 //# sourceMappingURL=WhisperSTT.js.map