Streaming STT - RunAnywhere Documentation

transcribeStream

for await (const event of RunAnywhere.stt.transcribeStream(audioIterable)) {
  if (event.type === 'partial') setPreview(event.text)
  if (event.type === 'final') setTranscript((t) => t + event.transcription.text)
}

Signature

transcribeStream(audio: AsyncIterable<AudioInput>, options?: SttOptions): AsyncIterable<TranscriptionEvent>

Building the audio iterable

Anything that yields AudioInput values works. From a recorder that hands you buffers:

import { AudioInputs } from '@runanywhere/core'

async function* microphone(): AsyncGenerator<AudioInput> {
  const queue: Uint8Array[] = []
  let notify: (() => void) | null = null

  recorder.on('data', (chunk: Uint8Array) => {
    queue.push(chunk)
    notify?.()
    notify = null
  })

  while (recording) {
    if (queue.length === 0) {
      await new Promise<void>((resolve) => {
        notify = resolve
      })
      continue
    }
    yield AudioInputs.pcm16(queue.shift()!)
  }
}

for await (const event of RunAnywhere.stt.transcribeStream(microphone())) {
  if (event.type === 'partial') setPreview(event.text)
}

The queue-and-notify pattern is what bridges a callback-based recorder to an async generator without dropping frames.

Stopping

Break out of the loop, then stop the recorder:

for await (const event of RunAnywhere.stt.transcribeStream(microphone())) {
  if (cancelled.current) break
  // …
}

recorder.stop()

AudioCaptureManager

@runanywhere/core also exports AudioCaptureManager and AudioPlaybackManager, which wrap the native audio path used by the voice session. If you are building a live transcription screen rather than a full conversation, they save you writing the bridge above.