@bottlenose/rxtranscribe/build/operators/toDeepSpeech.js

👂 Realtime speech-to-text (S2T) transcription with RxJS

(function (global, factory) {
  if (typeof define === "function" && define.amd) {
    define(["exports", "deepspeech", "rxjs", "rxjs/operators"], factory);
  } else if (typeof exports !== "undefined") {
    factory(exports, require("deepspeech"), require("rxjs"), require("rxjs/operators"));
  } else {
    var mod = {
      exports: {}
    };
    factory(mod.exports, global.deepspeech, global.rxjs, global.operators);
    global.toDeepSpeech = mod.exports;
  }
})(typeof globalThis !== "undefined" ? globalThis : typeof self !== "undefined" ? self : this, function (_exports, _deepspeech, _rxjs, _operators) {
  "use strict";

  Object.defineProperty(_exports, "__esModule", {
    value: true
  });
  _exports.default = _exports.testExports = void 0;
  const errors = {
    missingModelDir: () => new Error('modelDir<String> is required for toDeepSpeech operator')
  };

  const createModel = ({
    modelDir,
    beamWidth = 1024,
    lmAlpha = 0.75,
    lmBeta = 1.85,
    _newModel = dir => new _deepspeech.Model(dir)
  }) => {
    const modelPath = `${modelDir}.pbmm`;
    const scorerPath = `${modelDir}.scorer`;

    const newModel = _newModel(modelPath);

    newModel.setBeamWidth(beamWidth);
    newModel.enableExternalScorer(scorerPath);
    return newModel;
  };

  const transcribeChunks = ({
    model,
    sampleRate,
    candidateCount = 1
  }) => chunks => {
    const fullBuffer = Buffer.concat(chunks);
    const output = model.sttWithMetadata(fullBuffer, candidateCount);
    return output;
  };

  const transcribe = ({
    model,
    sampleRate = 16000,
    candidateCount = 1
  }) => bufferedChunks$ => bufferedChunks$.pipe((0, _operators.filter)(chunks => !!chunks && chunks.length), (0, _operators.map)(transcribeChunks({
    model,
    sampleRate,
    candidateCount
  })));

  const transcriptWordReducer = ({
    words,
    nextWord
  }, {
    text,
    start_time
  }) => text === ' ' ? {
    words: [...words, { ...nextWord,
      endTime: start_time
    }],
    nextWord: {
      text: '',
      startTime: null
    }
  } : {
    words,
    nextWord: {
      text: `${nextWord.text}${text}`,
      startTime: nextWord.startTime || start_time
    }
  };

  const standardizeOutput = () => deepSpeechMetadata => ({
    transcripts: deepSpeechMetadata.transcripts.map(transcript => ({
      confidence: transcript.confidence,
      words: transcript.tokens.reduce(transcriptWordReducer, {
        words: [],
        nextWord: {
          text: '',
          startTime: null
        }
      }).words
    }))
  });

  const toDeepSpeech = ({
    modelDir = process.env.DEEPSPEECH_MODEL_PATH,
    sampleRate = 16000,
    bufferSize = 3,
    candidateCount = 1,
    rawOutput = false,
    _createModel = createModel,
    _transcribe = transcribe
  } = {}) => {
    if (!modelDir) return () => (0, _rxjs.throwError)(errors.missingModelDir());

    const model = _createModel({
      modelDir
    });

    return fileChunk$ => fileChunk$.pipe((0, _operators.bufferCount)(bufferSize), _transcribe({
      model,
      sampleRate,
      candidateCount
    }), (0, _operators.map)(rawOutput ? metadata => metadata : standardizeOutput()));
  };

  const testExports = {
    createModel,
    standardizeOutput,
    transcribe,
    transcribeChunks
  };
  _exports.testExports = testExports;
  var _default = toDeepSpeech;
  _exports.default = _default;
});
//# sourceMappingURL=toDeepSpeech.js.map