STT + Cloud Function

​
This implementation is not suitable for long files as google cloud function times out after 1min to 9min ​
But for shorter files, you can add your media to a google cloud storage and run google cloud STT inside a cloud function and save the result to firestore.
​Firestore events​
1
// Goolge App Engine calls STT SDK
2
// this function returns null.
3
exports.createTranscript = functions.firestore
4
  .document("transcripts/{transcriptId}")
5
  .onCreate(async (change, context) => {
6
    // Get an object representing the document
7
    const newValue = change.data();
8
    // for now only run STT for uploaded files
9
    // eg if running from STT then don't run this
10
    if (newValue.transcriptionType === "upload") {
11
      // access a particular field as you would any JS property
12
      const storageRef = newValue.storageRef;
13
      console.log("storageRef", storageRef);
14
      // https://firebase.google.com/docs/storage/admin/start
15
      const storage = admin.storage();
16
      // https://github.com/firebase/firebase-tools/issues/1573#issuecomment-517000981
17
      const bucket = storage.appInternal.options.storageBucket;
18
      // // const gcsUri = 'gs://my-bucket/audio.raw';
19
      const gcsUri = `gs://${bucket}/${storageRef}`;
20
      console.log("gcsUri", gcsUri);
21
      //////// STT ////////
22
      // Creates a client for STT
23
      const client = new speech.SpeechClient();
24
      const encoding = "mp3"; //'Encoding of the audio file, e.g. LINEAR16';
25
      const sampleRateHertz = "48000"; //16000;
26
      // const languageCode = newValue.language.value;
27
      const languageCode = "en-US"; //'BCP-47 language code, e.g. en-US';
28
​
29
      const config = {
30
        encoding: encoding,
31
        sampleRateHertz: sampleRateHertz,
32
        languageCode: languageCode
33
      };
34
​
35
      const audio = {
36
        uri: gcsUri
37
      };
38
​
39
      const request = {
40
        config: config,
41
        audio: audio
42
      };
43
​
44
      //   // Detects speech in the audio file. This creates a recognition job that you
45
      //   // can wait for now, or get its result later.
46
      const [operation] = await client.longRunningRecognize(request);
47
      // Get a Promise representation of the final result of the job
48
      const [response] = await operation.promise();
49
      // TODO: Convert to DPE
50
      // TODO: Save DPE json format to firebase as collection
51
​
52
      // get text - tmp
53
      const transcription = response.results
54
        .map(result => result.alternatives[0].transcript)
55
        .join("\n");
56
      console.log(`Transcription: ${transcription}`);
57
​
58
      // Then return a promise of a set operation to update the document
59
      return change.ref.set(
60
        {
61
          // TODO: change transcription to transcript
62
          transcription: transcription
63
        },
64
        { merge: true }
65
      );
66
    } else {
67
      return null;
68
    }
69
  });
Copied!
links 
​https://github.com/GoogleCloudPlatform/nodejs-docs-samples/tree/master/functions/speech-to-speech​
​Supercharging Firebase with Google Cloud Platform - Google I/O 2016​
​A Serverless Audio Transcription Pipeline - audio-2-text​
​Transcribing videos​
​SpeechClient - SDK​
​REST Resource: operations​
​LongRunningRecognizeMetadata​
​LongRunningRecognizeResponse​
​Method: speech.recognize​

STT longRunningRecognize in Cloud function

STT + Google App Engine

Last modified 2yr ago

Copy link

Contents

links