https://github.com/rxtoolkit/stt-deepgram

👂 RxJS operator for realtime speech-to-text (STT/S2T) using Deepgram speeh-to-text
https://github.com/rxtoolkit/stt-deepgram

ai fp functional-programming observables package reactive-programming rxjs speech-to-text stt transcription

Last synced: 6 days ago
JSON representation

👂 RxJS operator for realtime speech-to-text (STT/S2T) using Deepgram speeh-to-text

• Host: GitHub
• URL: https://github.com/rxtoolkit/stt-deepgram
• Owner: rxtoolkit
• License: mit
• Created: 2021-02-09T21:43:59.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2024-02-16T18:39:24.000Z (11 months ago)
• Last Synced: 2024-12-10T01:59:30.047Z (about 1 month ago)
• Topics: ai, fp, functional-programming, observables, package, reactive-programming, rxjs, speech-to-text, stt, transcription
• Language: JavaScript
• Homepage:
• Size: 3.45 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 9
• Metadata Files:
  • Readme: README.md
  • Changelog: CHANGELOG.md
  • License: LICENSE.md

Awesome Lists containing this project

README

        
# @rxtk/stt-deepgram

> 👂 An RxJS operator for real-time speech-to-text (STT/S2T) streaming using the Deepgram speech-to-text API.

```bash

npm i @rxtk/stt-deepgram

```

```bash

yarn add @rxtk/stt-deepgram

```

⚠️ To run the Deepgram transcription pipeline, you will need to provide a deepgram username and password.

⚠️ node.js only. This has not been tested on Browsers but it might be possible to make it work.  If you get it working, please make a PR!

## Demo

Create a .env file with your `DEEPGRAM_PASSWORD` and `DEEPGRAM_USERNAME`:

```bash

echo 'DEEPGRAM_PASSWORD=MYPASSWORD' >> .env

echo 'DEEPGRAM_USERNAME=MYUSERNAME' >> .env

```

Run the demo!

```bash

yarn build # you must build the code before running the demo

yarn demo run

```

## API

### `toDeepgram`

Stream audio speech data to transcription service via WebSocket and get transcripts back:

```js

import {map} from 'rxjs/operators';

import {toDeepgram} from '@rxtk/stt-deepgram';

// The pipeline can take a stream of audio chunks encoded as 

// LINEAR16 (PCM encoded as 16-bit integers) in the form of a Buffer

const buffer$ = pcmChunkEncodedAs16BitIntegers$.pipe(

  map(chunk => Buffer.from(chunk, 'base64')),

  toDeepgram({

    username: process.env.DEEPGRAM_USERNAME,

    password: process.env.DEEPGRAM_PASSWORD,

  })

);

buffer$.subscribe(console.log); // log transcript output

```

> ⚠️ Pay attention to the endcoding of the audio data.  The operator only accepts PCM data encoded as 16-bit integers. For example, LINEAR16 encoding usually works.

## Guides

- [Introduction to audio data](https://developer.mozilla.org/en-US/docs/Web/Media/Formats/Audio_concepts)