Transcribe streaming audio from a microphone in TypeScript

Learn how to transcribe streaming audio in TypeScript.

Overview

By the end of this tutorial, you’ll be able to transcribe audio from your microphone in TypeScript.

Streaming Speech-to-Text is only available for English. See Supported languages.

Before you begin

To complete this tutorial, you need:

• Node.js installed. You can check to see if it is installed with node -v.
• TypeScript installed. You can check to see if it is installed with tsc -v
• An with credit card set up.

Here’s the full sample code for what you’ll build in this tutorial:

import { Readable } from 'stream'
import { AssemblyAI, RealtimeTranscript } from 'assemblyai'
import { SoxRecording } from './sox.js'

const run = async () => {
  const client = new AssemblyAI({
    apiKey: 'YOUR_API_KEY'
  })

  const SAMPLE_RATE = 16_000

  const transcriber = client.realtime.transcriber({
    sampleRate: SAMPLE_RATE
  })

  transcriber.on('open', ({ sessionId }) => {
    console.log(`Session opened with ID: ${sessionId}`)
  })

  transcriber.on('error', (error: Error) => {
    console.error('Error:', error)
  })

  transcriber.on('close', (code: number, reason: string) =>
    console.log('Session closed:', code, reason)
  )

  transcriber.on('transcript', (transcript: RealtimeTranscript) => {
    if (!transcript.text) {
      return
    }

    if (transcript.message_type === 'PartialTranscript') {
      console.log('Partial:', transcript.text)
    } else {
      console.log('Final:', transcript.text)
    }
  })

  console.log('Connecting to real-time transcript service')
  await transcriber.connect()

  console.log('Starting recording')
  const recording = new SoxRecording({
    channels: 1,
    sampleRate: SAMPLE_RATE,
    audioType: 'wav' // Linear PCM
  })

  recording.stream().pipeTo(transcriber.stream())

  // Stop recording and close connection using Ctrl-C.
  process.on('SIGINT', async function () {
    console.log()
    console.log('Stopping recording')
    recording.stop()

    console.log('Closing real-time transcript connection')
    await transcriber.close()

    process.exit()
  })
}

run()

Step 1: Install the SDK

Run npm init to create an NPM package, and then install the AssemblyAI package via NPM:

npm install assemblyai

Step 2: Configure the API key

In this step, you’ll create an SDK client and configure it to use your API key.

1. Browse to Account, and then click the text under Your API key to copy it.

2. Configure the SDK to use your API key. Create a file called main.ts and add the below code, replacing YOUR_API_KEY with your copied API key.

   import { AssemblyAI } from 'assemblyai'const client = new AssemblyAI({  apiKey: 'YOUR_API_KEY' })
   

Step 3: Create a streaming service

1. Create a new streaming service from the AssemblyAI client. If you don’t set a sample rate, it defaults to 16 kHz.

   const SAMPLE_RATE = 16_000const transcriber = client.realtime.transcriber({  sampleRate: SAMPLE_RATE})
   

   The sample_rate is the number of audio samples per second, measured in hertz (Hz). Higher sample rates result in higher quality audio, which may lead to better transcripts, but also more data being sent over the network.

   We recommend the following sample rates:

   • Minimum quality: 8_000 (8 kHz)
   • Medium quality: 16_000 (16 kHz)
   • Maximum quality: 48_000 (48 kHz)
2. Create functions to handle events from the real-time service.

   transcriber.on('open', ({ sessionId }) => {  console.log(`Session opened with ID: ${sessionId}`)})transcriber.on('error', (error: Error) => {  console.error('Error:', error)})transcriber.on('close', (code: number, reason: string) => {  console.log('Session closed:', code, reason)})
   
3. Create another function to handle transcripts. The real-time transcriber returns two types of transcripts: partial and final.

   • Partial transcripts are returned as the audio is being streamed to AssemblyAI.
   • Final transcripts are returned when the service detects a pause in speech.

   You can configure the silence threshold for automatic utterance detection and programmatically force the end of an utterance to immediately get a Final transcript.

   transcriber.on('transcript', (transcript: RealtimeTranscript) => {  if (!transcript.text) {    return  }  if (transcript.message_type === 'PartialTranscript') {    console.log('Partial:', transcript.text)  } else {    console.log('Final:', transcript.text)  }})
   

   You can also use the on("transcript.partial"), and on("transcript.final") callbacks to handle partial and final transcripts separately.

Step 4: Connect the streaming service

Streaming Speech-to-Text uses WebSockets to stream audio to AssemblyAI. This requires first establishing a connection to the API.

await transcriber.connect()

Step 5: Record audio from microphone

In this step, you’ll use SoX, a cross-platform audio library, to record audio from your microphone.

1. Install SoX on your machine.

   macOs

   brew install sox
   
2. Download the sox.ts (or sox.js) script to the root of your project, and import the SoxRecording class at the top of index.ts.

   import { SoxRecording } from './sox.js'
   

   The SoxRecording class lets you interact with SoX more easily.

3. In the on("open") callback, create a new microphone stream. The sampleRate needs to be the same value as the real-time service settings.

   const recording = new SoxRecording({
      channels: 1,
      sampleRate: SAMPLE_RATE,
      audioType: 'wav'
   })
   

   The SoxRecording formats the audio data for you. If you want to stream data from elsewhere, make sure that your audio data is in the following format:

   • Single channel
   • 16-bit signed integer PCM or mu-law encoding

   By default, the Streaming STT service expects PCM16-encoded audio. If you want to use mu-law encoding, see Specifying the encoding.

4. Pipe the recording stream to the real-time stream to send the audio for transcription.

   recording.stream().pipeTo(transcriber.stream())
   

   If you don’t use streams, you can also send buffers of audio data using transcriber.sendAudio(buffer).

Step 6: Disconnect the real-time service

When you are done, disconnect the transcriber to close the connection.

process.on('SIGINT', async function () {
  console.log()
  console.log('Stopping recording')
  recording.stop()

  console.log('Closing real-time transcript connection')
  await transcriber.close()

  process.exit()
})

To run the program, use tsc main.ts to compile the JavaScript file, and then run node main.js to run it.

Next steps

To learn more about Streaming Speech-to-Text, see the following resources:

• Streaming Speech-to-Text
• Streaming STT WebSocket API reference

Need some help?

If you get stuck, or have any other questions, we’d love to help you out. Ask our support team in our Discord server.