react-native-executorch 0.5.1-rc.0 → 0.5.1

package/lib/controllers/SpeechToTextController.js

@@ -1,343 +0,0 @@
-import { HAMMING_DIST_THRESHOLD, MODEL_CONFIGS, SECOND, MODES, NUM_TOKENS_TO_TRIM, STREAMING_ACTION, } from '../constants/sttDefaults';
-import { TokenizerModule } from '../modules/natural_language_processing/TokenizerModule';
-import { ResourceFetcher } from '../utils/ResourceFetcher';
-import { longCommonInfPref } from '../utils/stt';
-import { ETError, getError } from '../Error';
-import { Logger } from '../common/Logger';
-export class SpeechToTextController {
-    speechToTextNativeModule;
-    sequence = [];
-    isReady = false;
-    isGenerating = false;
-    tokenizerModule;
-    overlapSeconds;
-    windowSize;
-    chunks = [];
-    seqs = [];
-    prevSeq = [];
-    waveform = [];
-    numOfChunks = 0;
-    streaming = false;
-    // User callbacks
-    decodedTranscribeCallback;
-    modelDownloadProgressCallback;
-    isReadyCallback;
-    isGeneratingCallback;
-    onErrorCallback;
-    config;
-    constructor({ transcribeCallback, modelDownloadProgressCallback, isReadyCallback, isGeneratingCallback, onErrorCallback, overlapSeconds, windowSize, streamingConfig, }) {
-        this.tokenizerModule = new TokenizerModule();
-        this.decodedTranscribeCallback = async (seq) => transcribeCallback(await this.tokenIdsToText(seq));
-        this.modelDownloadProgressCallback = modelDownloadProgressCallback;
-        this.isReadyCallback = (isReady) => {
-            this.isReady = isReady;
-            isReadyCallback?.(isReady);
-        };
-        this.isGeneratingCallback = (isGenerating) => {
-            this.isGenerating = isGenerating;
-            isGeneratingCallback?.(isGenerating);
-        };
-        this.onErrorCallback = (error) => {
-            if (onErrorCallback) {
-                onErrorCallback(error ? new Error(getError(error)) : undefined);
-                return;
-            }
-            else {
-                throw new Error(getError(error));
-            }
-        };
-        this.configureStreaming(overlapSeconds, windowSize, streamingConfig || 'balanced');
-    }
-    async loadModel(modelName, encoderSource, decoderSource, tokenizerSource) {
-        this.onErrorCallback(undefined);
-        this.isReadyCallback(false);
-        this.config = MODEL_CONFIGS[modelName];
-        try {
-            await this.tokenizerModule.load(tokenizerSource || this.config.tokenizer.source);
-            const paths = await ResourceFetcher.fetch(this.modelDownloadProgressCallback, encoderSource || this.config.sources.encoder, decoderSource || this.config.sources.decoder);
-            if (paths === null || paths.length < 2) {
-                throw new Error('Download interrupted.');
-            }
-            [encoderSource, decoderSource] = paths;
-        }
-        catch (e) {
-            this.onErrorCallback(e);
-            return;
-        }
-        if (modelName === 'whisperMultilingual') {
-            // The underlying native class is instantiated based on the name of the model. There is no need to
-            // create a separate class for multilingual version of Whisper, since it is the same. We just need
-            // the distinction here, in TS, for start tokens and such. If we introduce
-            // more versions of Whisper, such as the small one, this should be refactored.
-            modelName = 'whisper';
-        }
-        try {
-            const nativeSpeechToText = await global.loadSpeechToText(encoderSource, decoderSource, modelName);
-            this.speechToTextNativeModule = nativeSpeechToText;
-            this.isReadyCallback(true);
-        }
-        catch (e) {
-            this.onErrorCallback(e);
-        }
-    }
-    configureStreaming(overlapSeconds, windowSize, streamingConfig) {
-        if (streamingConfig) {
-            this.windowSize = MODES[streamingConfig].windowSize * SECOND;
-            this.overlapSeconds = MODES[streamingConfig].overlapSeconds * SECOND;
-        }
-        if (streamingConfig && (windowSize || overlapSeconds)) {
-            Logger.warn(`windowSize and overlapSeconds overrides values from streamingConfig ${streamingConfig}.`);
-        }
-        this.windowSize = (windowSize || 0) * SECOND || this.windowSize;
-        this.overlapSeconds = (overlapSeconds || 0) * SECOND || this.overlapSeconds;
-        if (2 * this.overlapSeconds + this.windowSize >= 30 * SECOND) {
-            Logger.warn(`Invalid values for overlapSeconds and/or windowSize provided. Expected windowSize + 2 * overlapSeconds (== ${this.windowSize + 2 * this.overlapSeconds}) <= 30. Setting windowSize to ${30 * SECOND - 2 * this.overlapSeconds}.`);
-            this.windowSize = 30 * SECOND - 2 * this.overlapSeconds;
-        }
-    }
-    chunkWaveform() {
-        this.numOfChunks = Math.ceil(this.waveform.length / this.windowSize);
-        for (let i = 0; i < this.numOfChunks; i++) {
-            let chunk = [];
-            const left = Math.max(this.windowSize * i - this.overlapSeconds, 0);
-            const right = Math.min(this.windowSize * (i + 1) + this.overlapSeconds, this.waveform.length);
-            chunk = this.waveform.slice(left, right);
-            this.chunks.push(chunk);
-        }
-    }
-    resetState() {
-        this.sequence = [];
-        this.seqs = [];
-        this.waveform = [];
-        this.prevSeq = [];
-        this.chunks = [];
-        this.decodedTranscribeCallback([]);
-        this.onErrorCallback(undefined);
-    }
-    expectedChunkLength() {
-        //only first chunk can be of shorter length, for first chunk there are no seqs decoded
-        return this.seqs.length
-            ? this.windowSize + 2 * this.overlapSeconds
-            : this.windowSize + this.overlapSeconds;
-    }
-    async getStartingTokenIds(audioLanguage) {
-        // We need different starting token ids based on the multilingualism of the model.
-        // The eng version only needs BOS token, while the multilingual one needs:
-        // [BOS, LANG, TRANSCRIBE]. Optionally we should also set notimestamps token, as timestamps
-        // is not yet supported.
-        if (!audioLanguage) {
-            return [this.config.tokenizer.bos];
-        }
-        // FIXME: I should use .getTokenId for the BOS as well, should remove it from config
-        const langTokenId = await this.tokenizerModule.tokenToId(`<|${audioLanguage}|>`);
-        const transcribeTokenId = await this.tokenizerModule.tokenToId('<|transcribe|>');
-        const noTimestampsTokenId = await this.tokenizerModule.tokenToId('<|notimestamps|>');
-        const startingTokenIds = [
-            this.config.tokenizer.bos,
-            langTokenId,
-            transcribeTokenId,
-            noTimestampsTokenId,
-        ];
-        return startingTokenIds;
-    }
-    async decodeChunk(chunk, audioLanguage) {
-        const seq = await this.getStartingTokenIds(audioLanguage);
-        let prevSeqTokenIdx = 0;
-        this.prevSeq = this.sequence.slice();
-        try {
-            await this.encode(new Float32Array(chunk));
-        }
-        catch (error) {
-            this.onErrorCallback(new Error(getError(error) + ' encoding error'));
-            return [];
-        }
-        let lastToken = seq.at(-1);
-        while (lastToken !== this.config.tokenizer.eos) {
-            try {
-                lastToken = await this.decode(seq);
-            }
-            catch (error) {
-                this.onErrorCallback(new Error(getError(error) + ' decoding error'));
-                return [...seq, this.config.tokenizer.eos];
-            }
-            seq.push(lastToken);
-            if (this.seqs.length > 0 &&
-                seq.length < this.seqs.at(-1).length &&
-                seq.length % 3 !== 0) {
-                this.prevSeq.push(this.seqs.at(-1)[prevSeqTokenIdx++]);
-                this.decodedTranscribeCallback(this.prevSeq);
-            }
-        }
-        return seq;
-    }
-    async handleOverlaps(seqs) {
-        const maxInd = longCommonInfPref(seqs.at(-2), seqs.at(-1), HAMMING_DIST_THRESHOLD);
-        this.sequence = [...this.sequence, ...seqs.at(-2).slice(0, maxInd)];
-        this.decodedTranscribeCallback(this.sequence);
-        return this.sequence.slice();
-    }
-    trimLeft(numOfTokensToTrim) {
-        const idx = this.seqs.length - 1;
-        if (this.seqs[idx][0] === this.config.tokenizer.bos) {
-            this.seqs[idx] = this.seqs[idx].slice(numOfTokensToTrim);
-        }
-    }
-    trimRight(numOfTokensToTrim) {
-        const idx = this.seqs.length - 2;
-        if (this.seqs[idx].at(-1) === this.config.tokenizer.eos) {
-            this.seqs[idx] = this.seqs[idx].slice(0, -numOfTokensToTrim);
-        }
-    }
-    // since we are calling this every time (except first) after a new seq is pushed to this.seqs
-    // we can only trim left the last seq and trim right the second to last seq
-    async trimSequences(audioLanguage) {
-        const numSpecialTokens = (await this.getStartingTokenIds(audioLanguage))
-            .length;
-        this.trimLeft(numSpecialTokens + NUM_TOKENS_TO_TRIM);
-        this.trimRight(numSpecialTokens + NUM_TOKENS_TO_TRIM);
-    }
-    // if last chunk is too short combine it with second to last to improve quality
-    validateAndFixLastChunk() {
-        if (this.chunks.length < 2)
-            return;
-        const lastChunkLength = this.chunks.at(-1).length / SECOND;
-        const secondToLastChunkLength = this.chunks.at(-2).length / SECOND;
-        if (lastChunkLength < 5 && secondToLastChunkLength + lastChunkLength < 30) {
-            this.chunks[this.chunks.length - 2] = [
-                ...this.chunks.at(-2).slice(0, -this.overlapSeconds * 2),
-                ...this.chunks.at(-1),
-            ];
-            this.chunks = this.chunks.slice(0, -1);
-        }
-    }
-    async tokenIdsToText(tokenIds) {
-        try {
-            return await this.tokenizerModule.decode(tokenIds, true);
-        }
-        catch (e) {
-            this.onErrorCallback(new Error(`An error has occurred when decoding the token ids: ${e}`));
-            return '';
-        }
-    }
-    async transcribe(waveform, audioLanguage) {
-        try {
-            if (!this.isReady)
-                throw Error(getError(ETError.ModuleNotLoaded));
-            if (this.isGenerating || this.streaming)
-                throw Error(getError(ETError.ModelGenerating));
-            if (!!audioLanguage !== this.config.isMultilingual)
-                throw new Error(getError(ETError.MultilingualConfiguration));
-        }
-        catch (e) {
-            this.onErrorCallback(e);
-            return '';
-        }
-        // Making sure that the error is not set when we get there
-        this.isGeneratingCallback(true);
-        this.resetState();
-        this.waveform = waveform;
-        this.chunkWaveform();
-        this.validateAndFixLastChunk();
-        for (let chunkId = 0; chunkId < this.chunks.length; chunkId++) {
-            const seq = await this.decodeChunk(this.chunks.at(chunkId), audioLanguage);
-            // whole audio is inside one chunk, no processing required
-            if (this.chunks.length === 1) {
-                this.sequence = seq;
-                this.decodedTranscribeCallback(seq);
-                break;
-            }
-            this.seqs.push(seq);
-            if (this.seqs.length < 2)
-                continue;
-            // Remove starting tokenIds and some additional ones
-            await this.trimSequences(audioLanguage);
-            this.prevSeq = await this.handleOverlaps(this.seqs);
-            // last sequence processed
-            // overlaps are already handled, so just append the last seq
-            if (this.seqs.length === this.chunks.length) {
-                this.sequence = [...this.sequence, ...this.seqs.at(-1)];
-                this.decodedTranscribeCallback(this.sequence);
-                this.prevSeq = this.sequence;
-            }
-        }
-        const decodedText = await this.tokenIdsToText(this.sequence);
-        this.isGeneratingCallback(false);
-        return decodedText;
-    }
-    async streamingTranscribe(streamAction, waveform, audioLanguage) {
-        try {
-            if (!this.isReady)
-                throw Error(getError(ETError.ModuleNotLoaded));
-            if (!!audioLanguage !== this.config.isMultilingual)
-                throw new Error(getError(ETError.MultilingualConfiguration));
-            if (streamAction === STREAMING_ACTION.START &&
-                !this.streaming &&
-                this.isGenerating)
-                throw Error(getError(ETError.ModelGenerating));
-            if (streamAction === STREAMING_ACTION.START && this.streaming)
-                throw Error(getError(ETError.ModelGenerating));
-            if (streamAction === STREAMING_ACTION.DATA && !this.streaming)
-                throw Error(getError(ETError.StreamingNotStarted));
-            if (streamAction === STREAMING_ACTION.STOP && !this.streaming)
-                throw Error(getError(ETError.StreamingNotStarted));
-            if (streamAction === STREAMING_ACTION.DATA && !waveform)
-                throw new Error(getError(ETError.MissingDataChunk));
-        }
-        catch (e) {
-            this.onErrorCallback(e);
-            return '';
-        }
-        if (streamAction === STREAMING_ACTION.START) {
-            this.resetState();
-            this.streaming = true;
-            this.isGeneratingCallback(true);
-        }
-        this.waveform = [...this.waveform, ...(waveform || [])];
-        // while buffer has at least required size get chunk and decode
-        while (this.waveform.length >= this.expectedChunkLength()) {
-            const chunk = this.waveform.slice(0, this.windowSize +
-                this.overlapSeconds * (1 + Number(this.seqs.length > 0)));
-            this.chunks = [chunk]; //save last chunk for STREAMING_ACTION.STOP
-            this.waveform = this.waveform.slice(this.windowSize - this.overlapSeconds * Number(this.seqs.length === 0));
-            const seq = await this.decodeChunk(chunk, audioLanguage);
-            this.seqs.push(seq);
-            if (this.seqs.length < 2)
-                continue;
-            await this.trimSequences(audioLanguage);
-            await this.handleOverlaps(this.seqs);
-        }
-        // got final package, process all remaining waveform data
-        // since we run the loop above the waveform has at most one chunk in it
-        if (streamAction === STREAMING_ACTION.STOP) {
-            // pad remaining waveform data with previous chunk to this.windowSize + 2 * this.overlapSeconds
-            const chunk = this.chunks.length
-                ? [
-                    ...this.chunks[0].slice(0, this.windowSize),
-                    ...this.waveform,
-                ].slice(-this.windowSize - 2 * this.overlapSeconds)
-                : this.waveform;
-            this.waveform = [];
-            const seq = await this.decodeChunk(chunk, audioLanguage);
-            this.seqs.push(seq);
-            if (this.seqs.length === 1) {
-                this.sequence = this.seqs[0];
-            }
-            else {
-                await this.trimSequences(audioLanguage);
-                await this.handleOverlaps(this.seqs);
-                this.sequence = [...this.sequence, ...this.seqs.at(-1)];
-            }
-            this.decodedTranscribeCallback(this.sequence);
-            this.isGeneratingCallback(false);
-            this.streaming = false;
-        }
-        const decodedText = await this.tokenIdsToText(this.sequence);
-        return decodedText;
-    }
-    async encode(waveform) {
-        return await this.speechToTextNativeModule.encode(waveform);
-    }
-    async decode(seq) {
-        return await this.speechToTextNativeModule.decode(seq);
-    }
-}

package/lib/hooks/natural_language_processing/useSpeechToText.js

@@ -1,44 +0,0 @@
-import { useEffect, useMemo, useState } from 'react';
-import { SpeechToTextController } from '../../controllers/SpeechToTextController';
-export const useSpeechToText = ({ modelName, encoderSource, decoderSource, tokenizerSource, overlapSeconds, windowSize, streamingConfig, preventLoad = false, }) => {
-    const [sequence, setSequence] = useState('');
-    const [isReady, setIsReady] = useState(false);
-    const [downloadProgress, setDownloadProgress] = useState(0);
-    const [isGenerating, setIsGenerating] = useState(false);
-    const [error, setError] = useState();
-    const model = useMemo(() => new SpeechToTextController({
-        transcribeCallback: setSequence,
-        isReadyCallback: setIsReady,
-        isGeneratingCallback: setIsGenerating,
-        onErrorCallback: setError,
-        modelDownloadProgressCallback: setDownloadProgress,
-    }), []);
-    useEffect(() => {
-        model.configureStreaming(overlapSeconds, windowSize, streamingConfig);
-    }, [model, overlapSeconds, windowSize, streamingConfig]);
-    useEffect(() => {
-        const loadModel = async () => {
-            await model.loadModel(modelName, encoderSource, decoderSource, tokenizerSource);
-        };
-        if (!preventLoad) {
-            loadModel();
-        }
-    }, [
-        model,
-        modelName,
-        encoderSource,
-        decoderSource,
-        tokenizerSource,
-        preventLoad,
-    ]);
-    return {
-        isReady,
-        isGenerating,
-        downloadProgress,
-        configureStreaming: model.configureStreaming,
-        sequence,
-        error,
-        transcribe: (waveform, audioLanguage) => model.transcribe(waveform, audioLanguage),
-        streamingTranscribe: (streamAction, waveform, audioLanguage) => model.streamingTranscribe(streamAction, waveform, audioLanguage),
-    };
-};

package/lib/index.d.ts

@@ -1,50 +0,0 @@
-import { SpeechToTextLanguage } from './types/stt';
-declare global {
-    var loadStyleTransfer: (source: string) => any;
-    var loadImageSegmentation: (source: string) => any;
-    var loadClassification: (source: string) => any;
-    var loadObjectDetection: (source: string) => any;
-    var loadExecutorchModule: (source: string) => any;
-    var loadTokenizerModule: (source: string) => any;
-    var loadImageEmbeddings: (source: string) => any;
-    var loadTextEmbeddings: (modelSource: string, tokenizerSource: string) => any;
-    var loadLLM: (modelSource: string, tokenizerSource: string) => any;
-    var loadSpeechToText: (encoderSource: string, decoderSource: string, modelName: string) => any;
-    var loadOCR: (detectorSource: string, recognizerLarge: string, recognizerMedium: string, recognizerSmall: string, symbols: string) => any;
-    var loadVerticalOCR: (detectorLarge: string, detectorNarrow: string, recognizer: string, symbols: string, independentCharacters?: boolean) => any;
-}
-export * from './hooks/computer_vision/useClassification';
-export * from './hooks/computer_vision/useObjectDetection';
-export * from './hooks/computer_vision/useStyleTransfer';
-export * from './hooks/computer_vision/useImageSegmentation';
-export * from './hooks/computer_vision/useOCR';
-export * from './hooks/computer_vision/useVerticalOCR';
-export * from './hooks/computer_vision/useImageEmbeddings';
-export * from './hooks/natural_language_processing/useLLM';
-export * from './hooks/natural_language_processing/useSpeechToText';
-export * from './hooks/natural_language_processing/useTextEmbeddings';
-export * from './hooks/natural_language_processing/useTokenizer';
-export * from './hooks/general/useExecutorchModule';
-export * from './modules/computer_vision/ClassificationModule';
-export * from './modules/computer_vision/ObjectDetectionModule';
-export * from './modules/computer_vision/StyleTransferModule';
-export * from './modules/computer_vision/ImageSegmentationModule';
-export * from './modules/computer_vision/OCRModule';
-export * from './modules/computer_vision/VerticalOCRModule';
-export * from './modules/general/ExecutorchModule';
-export * from './modules/computer_vision/ImageEmbeddingsModule';
-export * from './modules/natural_language_processing/LLMModule';
-export * from './modules/natural_language_processing/SpeechToTextModule';
-export * from './modules/natural_language_processing/TextEmbeddingsModule';
-export * from './modules/natural_language_processing/TokenizerModule';
-export * from './utils/ResourceFetcher';
-export * from './utils/llm';
-export * from './types/objectDetection';
-export * from './types/ocr';
-export * from './types/imageSegmentation';
-export * from './types/llm';
-export { SpeechToTextLanguage };
-export * from './constants/modelUrls';
-export * from './constants/ocr/models';
-export * from './constants/llmDefaults';
-export { STREAMING_ACTION, MODES } from './constants/sttDefaults';

package/lib/index.js

@@ -1,59 +0,0 @@
-import { SpeechToTextLanguage } from './types/stt';
-import { ETInstallerNativeModule } from './native/RnExecutorchModules';
-// eslint-disable no-var
-if (global.loadStyleTransfer == null ||
-    global.loadImageSegmentation == null ||
-    global.loadExecutorchModule == null ||
-    global.loadClassification == null ||
-    global.loadObjectDetection == null ||
-    global.loadTokenizerModule == null ||
-    global.loadTextEmbeddings == null ||
-    global.loadImageEmbeddings == null ||
-    global.loadLLM == null ||
-    global.loadSpeechToText == null ||
-    global.loadOCR == null) {
-    if (!ETInstallerNativeModule) {
-        throw new Error(`Failed to install react-native-executorch: The native module could not be found.`);
-    }
-    ETInstallerNativeModule.install();
-}
-// hooks
-export * from './hooks/computer_vision/useClassification';
-export * from './hooks/computer_vision/useObjectDetection';
-export * from './hooks/computer_vision/useStyleTransfer';
-export * from './hooks/computer_vision/useImageSegmentation';
-export * from './hooks/computer_vision/useOCR';
-export * from './hooks/computer_vision/useVerticalOCR';
-export * from './hooks/computer_vision/useImageEmbeddings';
-export * from './hooks/natural_language_processing/useLLM';
-export * from './hooks/natural_language_processing/useSpeechToText';
-export * from './hooks/natural_language_processing/useTextEmbeddings';
-export * from './hooks/natural_language_processing/useTokenizer';
-export * from './hooks/general/useExecutorchModule';
-// modules
-export * from './modules/computer_vision/ClassificationModule';
-export * from './modules/computer_vision/ObjectDetectionModule';
-export * from './modules/computer_vision/StyleTransferModule';
-export * from './modules/computer_vision/ImageSegmentationModule';
-export * from './modules/computer_vision/OCRModule';
-export * from './modules/computer_vision/VerticalOCRModule';
-export * from './modules/general/ExecutorchModule';
-export * from './modules/computer_vision/ImageEmbeddingsModule';
-export * from './modules/natural_language_processing/LLMModule';
-export * from './modules/natural_language_processing/SpeechToTextModule';
-export * from './modules/natural_language_processing/TextEmbeddingsModule';
-export * from './modules/natural_language_processing/TokenizerModule';
-// utils
-export * from './utils/ResourceFetcher';
-export * from './utils/llm';
-// types
-export * from './types/objectDetection';
-export * from './types/ocr';
-export * from './types/imageSegmentation';
-export * from './types/llm';
-export { SpeechToTextLanguage };
-// constants
-export * from './constants/modelUrls';
-export * from './constants/ocr/models';
-export * from './constants/llmDefaults';
-export { STREAMING_ACTION, MODES } from './constants/sttDefaults';

package/lib/module/constants/sttDefaults.js

@@ -1,74 +0,0 @@
-"use strict";
-
-import { MOONSHINE_TINY, WHISPER_TINY, WHISPER_TINY_MULTILINGUAL } from './modelUrls';
-import { AvailableModels } from '../types/stt';
-export const SAMPLE_RATE = 16_000;
-export const SECOND = SAMPLE_RATE;
-export const HAMMING_DIST_THRESHOLD = 1;
-const whisperTinyModelConfig = {
-  sources: {
-    encoder: WHISPER_TINY.encoderSource,
-    decoder: WHISPER_TINY.decoderSource
-  },
-  tokenizer: {
-    source: WHISPER_TINY.tokenizerSource,
-    bos: 50257,
-    // FIXME: this is a placeholder and needs to be changed
-    eos: 50256 // FIXME: this is a placeholder and needs to be changed
-  },
-  isMultilingual: false
-};
-const moonshineTinyModelConfig = {
-  sources: {
-    encoder: MOONSHINE_TINY.encoderSource,
-    decoder: MOONSHINE_TINY.decoderSource
-  },
-  tokenizer: {
-    source: MOONSHINE_TINY.tokenizerSource,
-    bos: 1,
-    // FIXME: this is a placeholder and needs to be changed
-    eos: 2 // FIXME: this is a placeholder and needs to be changed
-  },
-  isMultilingual: false
-};
-const whisperTinyMultilingualModelConfig = {
-  sources: {
-    encoder: WHISPER_TINY_MULTILINGUAL.encoderSource,
-    decoder: WHISPER_TINY_MULTILINGUAL.decoderSource
-  },
-  tokenizer: {
-    source: WHISPER_TINY_MULTILINGUAL.tokenizerSource,
-    bos: 50258,
-    // FIXME: this is a placeholder and needs to be changed
-    eos: 50257 // FIXME: this is a placeholder and needs to be changed
-  },
-  isMultilingual: true
-};
-export const MODEL_CONFIGS = {
-  moonshine: moonshineTinyModelConfig,
-  whisper: whisperTinyModelConfig,
-  whisperMultilingual: whisperTinyMultilingualModelConfig
-};
-export const MODES = {
-  fast: {
-    windowSize: 5,
-    overlapSeconds: 1.2
-  },
-  balanced: {
-    windowSize: 12,
-    overlapSeconds: 2
-  },
-  quality: {
-    windowSize: 24,
-    overlapSeconds: 3
-  }
-};
-export const NUM_TOKENS_TO_TRIM = 3;
-export let STREAMING_ACTION = /*#__PURE__*/function (STREAMING_ACTION) {
-  STREAMING_ACTION[STREAMING_ACTION["START"] = 0] = "START";
-  STREAMING_ACTION[STREAMING_ACTION["DATA"] = 1] = "DATA";
-  STREAMING_ACTION[STREAMING_ACTION["STOP"] = 2] = "STOP";
-  return STREAMING_ACTION;
-}({});
-export { AvailableModels };
-//# sourceMappingURL=sttDefaults.js.map

package/lib/module/constants/sttDefaults.js.map

@@ -1 +0,0 @@
-{"version":3,"names":["MOONSHINE_TINY","WHISPER_TINY","WHISPER_TINY_MULTILINGUAL","AvailableModels","SAMPLE_RATE","SECOND","HAMMING_DIST_THRESHOLD","whisperTinyModelConfig","sources","encoder","encoderSource","decoder","decoderSource","tokenizer","source","tokenizerSource","bos","eos","isMultilingual","moonshineTinyModelConfig","whisperTinyMultilingualModelConfig","MODEL_CONFIGS","moonshine","whisper","whisperMultilingual","MODES","fast","windowSize","overlapSeconds","balanced","quality","NUM_TOKENS_TO_TRIM","STREAMING_ACTION"],"sourceRoot":"../../../src","sources":["constants/sttDefaults.ts"],"mappings":";;AAAA,SACEA,cAAc,EACdC,YAAY,EACZC,yBAAyB,QACpB,aAAa;AACpB,SAASC,eAAe,QAAqB,cAAc;AAE3D,OAAO,MAAMC,WAAW,GAAG,MAAM;AACjC,OAAO,MAAMC,MAAM,GAAGD,WAAW;AACjC,OAAO,MAAME,sBAAsB,GAAG,CAAC;AAEvC,MAAMC,sBAAsB,GAAG;EAC7BC,OAAO,EAAE;IACPC,OAAO,EAAER,YAAY,CAACS,aAAa;IACnCC,OAAO,EAAEV,YAAY,CAACW;EACxB,CAAC;EACDC,SAAS,EAAE;IACTC,MAAM,EAAEb,YAAY,CAACc,eAAe;IACpCC,GAAG,EAAE,KAAK;IAAE;IACZC,GAAG,EAAE,KAAK,CAAE;EACd,CAAC;EACDC,cAAc,EAAE;AAClB,CAAC;AAED,MAAMC,wBAAwB,GAAG;EAC/BX,OAAO,EAAE;IACPC,OAAO,EAAET,cAAc,CAACU,aAAa;IACrCC,OAAO,EAAEX,cAAc,CAACY;EAC1B,CAAC;EACDC,SAAS,EAAE;IACTC,MAAM,EAAEd,cAAc,CAACe,eAAe;IACtCC,GAAG,EAAE,CAAC;IAAE;IACRC,GAAG,EAAE,CAAC,CAAE;EACV,CAAC;EACDC,cAAc,EAAE;AAClB,CAAC;AAED,MAAME,kCAAkC,GAAG;EACzCZ,OAAO,EAAE;IACPC,OAAO,EAAEP,yBAAyB,CAACQ,aAAa;IAChDC,OAAO,EAAET,yBAAyB,CAACU;EACrC,CAAC;EACDC,SAAS,EAAE;IACTC,MAAM,EAAEZ,yBAAyB,CAACa,eAAe;IACjDC,GAAG,EAAE,KAAK;IAAE;IACZC,GAAG,EAAE,KAAK,CAAE;EACd,CAAC;EACDC,cAAc,EAAE;AAClB,CAAC;AAED,OAAO,MAAMG,aAEZ,GAAG;EACFC,SAAS,EAAEH,wBAAwB;EACnCI,OAAO,EAAEhB,sBAAsB;EAC/BiB,mBAAmB,EAAEJ;AACvB,CAAC;AAED,OAAO,MAAMK,KAAK,GAAG;EACnBC,IAAI,EAAE;IACJC,UAAU,EAAE,CAAC;IACbC,cAAc,EAAE;EAClB,CAAC;EACDC,QAAQ,EAAE;IACRF,UAAU,EAAE,EAAE;IACdC,cAAc,EAAE;EAClB,CAAC;EACDE,OAAO,EAAE;IACPH,UAAU,EAAE,EAAE;IACdC,cAAc,EAAE;EAClB;AACF,CAAC;AAED,OAAO,MAAMG,kBAAkB,GAAG,CAAC;AAEnC,WAAYC,gBAAgB,0BAAhBA,gBAAgB;EAAhBA,gBAAgB,CAAhBA,gBAAgB;EAAhBA,gBAAgB,CAAhBA,gBAAgB;EAAhBA,gBAAgB,CAAhBA,gBAAgB;EAAA,OAAhBA,gBAAgB;AAAA;AAM5B,SAAS7B,eAAe","ignoreList":[]}