Transcribe video/audio using IBM Watson

transcribe.js

var request = require('request');
var fs = require('fs');
var sox = require('sox');
var spawn = require('child_process').spawn;

var WATSON_USER = '';
var WATSON_PASS = '';
var url = 'https://stream.watsonplatform.net/speech-to-text/api/v1/recognize';



function convertToWav(file, cb) {
  var aud_file = file + '.temp.wav';
  fs.exists(aud_file, function(exists) {
    if (exists) {
      cb(aud_file);
    } else {
      var convert = spawn('ffmpeg', ['-i', file, '-acodec', 'pcm_s16le', '-ac', '1', '-ar', '16000', aud_file]);
      convert.on('close', function() {
        cb(aud_file);
      });
    }
  });
}


function split(file, cb) {
  //split into 5 minute chunks
  var maxLength = 60 * 5;
  var total = 0;
  var files = [];
  sox.identify(file, function(err, results) {
    var duration = results.duration;
    if (duration > maxLength) {
      var i = 0;
      while (i < duration) {
        total++;
        var filePart = file + '.' + i + '.wav';
        var args = [file, filePart, 'trim', '' + i, '' + maxLength];
        var s = spawn('sox', args);
        s.on('close', finishedSplit.bind({}, filePart, i));
        i += maxLength;
      }
    } else {
      cb([{
        name: filename,
        offset: 0
      }]);
    }
  });

  function finishedSplit(filename, start) {
    files.push({
      name: filename,
      offset: start
    });
    total--;
    if (total === 0) {
      cb(files);
    }
  }
}


function send_to_watson(file, callback) {
  fs.readFile(file, function(error, data) {
    if (error) {
      console.log(error);
      return false;
    }

    var options = {
      headers: {
        'Content-Type': 'audio/wav',
        'Transfer-Encoding': 'chunked',
      },
      qs: {
        timestamps: 'true',
        continuous: 'true',
        inactivity_timeout: '-1'
      },
      auth: {
        user: WATSON_USER,
        pass: WATSON_PASS
      },
      body: data
    };

    request.post(url, options, function(error, response, body) {
      if (error) console.log(error);

      // var parsed = parse(JSON.parse(body));
      if (typeof callback !== 'undefined') {
        callback(JSON.parse(body));
      }
    });
  });
}


function parse(data, offset) {
  if (typeof offset === 'undefined') offset = 0;

  var out = {
    sentences: []
  };

  out.sentences = data.results.map(function(r) {
    var item = {};
    var _item = r.alternatives[0];
    item.transcript = _item.transcript;
    item.confidence = _item.confidence;
    item.words = _item.timestamps.map(function(t) {
      return {
        word: t[0],
        start: t[1] + offset,
        end: t[2] + offset
      };
    });
    return item;
  });

  return out;
}


function writeOut(file, data) {
  var out = [];

  data = data.sort(function(a, b) {
    return a.f.offset - b.f.offset;
  });

  data.forEach(function(d) {
    out = out.concat(d.data.sentences);
  });

  fs.writeFile(file + '.transcription.json', JSON.stringify(out), function() {
    if (data.length > 1) {
      data.forEach(function(d) {
        fs.unlink(d.f.name);
      });
    }
  });
}


function transcribe(file, cb) {
  var out = [];
  var total = 0;

  convertToWav(file, function(newFile) {

    split(newFile, function(files) {
      total = files.length;

      files.forEach(function(f) {
        send_to_watson(f.name, function(data) {
          var parsed = parse(data, f.offset);
          out.push({
            f: f,
            data: parsed
          });
          total--;
          if (total === 0) {
            writeOut(file, out);
          }
        });

      });
    });
  });
}

var file = process.argv[2];

transcribe(file);

module.exports = transcribe;

You need sox & ffmpeg installed for this to work...