useAudioStream

React Hook for OpenAI Whisper API with speech recorder, real-time transcription and silence removal built-in

---

• Demo

• Real-Time transcription demo

https://user-images.githubusercontent.com/2707253/224465747-0b1ee159-21dd-4cd0-af9d-6fc9b882d716.mp4

---

• Announcement

  useAudioStream for React Native is being developed.

Repository: https://github.com/ElectricSpeech/use-audio-stream-native

Progress: https://github.com/ElectricSpeech/use-audio-stream-native/issues/1

• Install

npm i @dhansoo/use-audio-stream

yarn add @dhansoo/use-audio-stream

• Usage

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const {
5    recording,
6    speaking,
7    transcribing,
8    transcript,
9    pauseRecording,
10    startRecording,
11    stopRecording,
12  } = useAudioStream({
13    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
14  })
15
16  return (
17    <div>
18      <p>Recording: {recording}</p>
19      <p>Speaking: {speaking}</p>
20      <p>Transcribing: {transcribing}</p>
21      <p>Transcribed Text: {transcript.text}</p>
22      <button onClick={() => startRecording()}>Start</button>
23      <button onClick={() => pauseRecording()}>Pause</button>
24      <button onClick={() => stopRecording()}>Stop</button>
25    </div>
26  )
27}

• Custom Server (keep OpenAI API token secure)

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  /**
5   * you have more control like this
6   * do whatever you want with the recorded speech
7   * send it to your own custom server
8   * and return the response back to useAudioStream
9   */
10  const onTranscribe = (blob: Blob) => {
11    const base64 = await new Promise<string | ArrayBuffer | null>(
12      (resolve) => {
13        const reader = new FileReader()
14        reader.onloadend = () => resolve(reader.result)
15        reader.readAsDataURL(blob)
16      }
17    )
18    const body = JSON.stringify({ file: base64, model: 'whisper-1' })
19    const headers = { 'Content-Type': 'application/json' }
20    const { default: axios } = await import('axios')
21    const response = await axios.post('/api/whisper', body, {
22      headers,
23    })
24    const { text } = await response.data
25    // you must return result from your server in Transcript format
26    return {
27      blob,
28      text,
29    }
30  }
31
32  const { transcript } = useAudioStream({
33    // callback to handle transcription with custom server
34    onTranscribe,
35  })
36
37  return (
38    <div>
39      <p>{transcript.text}</p>
40    </div>
41  )
42}

• Examples

• Real-time streaming trascription

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const { transcript } = useAudioStream({
5    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
6    streaming: true,
7    timeSlice: 1_000, // 1 second
8    whisperConfig: {
9      language: 'en',
10    },
11  })
12
13  return (
14    <div>
15      <p>{transcript.text}</p>
16    </div>
17  )
18}

• Remove silence before sending to Whisper to save cost

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const { transcript } = useAudioStream({
5    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
6    // use ffmpeg-wasp to remove silence from recorded speech
7    removeSilence: true,
8  })
9
10  return (
11    <div>
12      <p>{transcript.text}</p>
13    </div>
14  )
15}

• Auto start recording on component mounted

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const { transcript } = useAudioStream({
5    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
6    // will auto start recording speech upon component mounted
7    autoStart: true,
8  })
9
10  return (
11    <div>
12      <p>{transcript.text}</p>
13    </div>
14  )
15}

• Keep recording as long as the user is speaking

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const { transcript } = useAudioStream({
5    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
6    nonStop: true, // keep recording as long as the user is speaking
7    stopTimeout: 5000, // auto stop after 5 seconds
8  })
9
10  return (
11    <div>
12      <p>{transcript.text}</p>
13    </div>
14  )
15}

• Customize Whisper API config when autoTranscribe is true

1import { useAudioStream } from '@dhansoo/use-audio-stream'
2
3const App = () => {
4  const { transcript } = useAudioStream({
5    apiKey: process.env.OPENAI_API_TOKEN, // YOUR_OPEN_AI_TOKEN
6    autoTranscribe: true,
7    whisperConfig: {
8      prompt: 'previous conversation', // you can pass previous conversation for context
9      response_format: 'text', // output text instead of json
10      temperature: 0.8, // random output
11      language: 'es', // Spanish
12    },
13  })
14
15  return (
16    <div>
17      <p>{transcript.text}</p>
18    </div>
19  )
20}

• Dependencies

  • @dhansoo/react-hooks-async asynchronous react hooks
  • recordrtc: cross-browser audio recorder
  • lamejs encode wav into mp3 for cross-browser support
  • @ffmpeg/ffmpeg: for silence removal feature
  • hark: for speaking detection
  • axios: since fetch does not work with Whisper endpoint

most of these dependecies are lazy loaded, so it is only imported when it is needed

• API

• Config Object

Name	Type	Default Value	Description
apiKey	string	''	your OpenAI API token
autoStart	boolean	false	auto start speech recording on component mount
autoTranscribe	boolean	true	should auto transcribe after stop recording
mode	string	transcriptions	control Whisper mode either transcriptions or translations, currently only support translation to English
nonStop	boolean	false	if true, record will auto stop after stopTimeout. However if user keep on speaking, the recorder will keep recording
removeSilence	boolean	false	remove silence before sending file to OpenAI API
stopTimeout	number	5,000 ms	if nonStop is true, this become required. This control when the recorder auto stop
streaming	boolean	false	transcribe speech in real-time based on timeSlice
timeSlice	number	1000 ms	interval between each onDataAvailable event
whisperConfig	WhisperApiConfig	undefined	Whisper API transcription config
onDataAvailable	(blob: Blob) => void	undefined	callback function for getting recorded blob in interval between timeSlice
onTranscribe	(blob: Blob) => Promise<Transcript>	undefined	callback function to handle transcription on your own custom server

• WhisperApiConfig

Name	Type	Default Value	Description
prompt	string	undefined	An optional text to guide the model's style or continue a previous audio segment. The prompt should match the audio language.
response_format	string	json	The format of the transcript output, in one of these options: json, text, srt, verbose_json, or vtt.
temperature	number	0	The sampling temperature, between 0 and 1. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. If set to 0, the model will use log probability to automatically increase the temperature until certain thresholds are hit.
language	string	en	The language of the input audio. Supplying the input language in ISO-639-1 format will improve accuracy and latency.

• Return Object

Name	Type	Description
recording	boolean	speech recording state
speaking	boolean	detect when user is speaking
transcribing	boolean	while removing silence from speech and send request to OpenAI Whisper API
transcript	Transcript	object return after Whisper transcription complete
pauseRecording	Promise	pause speech recording
startRecording	Promise	start speech recording
stopRecording	Promise	stop speech recording

• Transcript

Name	Type	Description
blob	Blob	recorded speech in JavaScript Blob
text	string	transcribed text returned from Whisper API

• Roadmap

  • react-native support, will be export as use-audio-stream/native

---

Contact me for web or mobile app development using React or React Native
https://ElectricSpeech.github.io