transcribe() - RunAnywhere Documentation

import { RunAnywhere, AudioInputs } from '@runanywhere/core'
import { ONNX } from '@runanywhere/onnx'

await ONNX.register()

const transcription = await RunAnywhere.stt.transcribe(AudioInputs.file(path))
console.log(transcription.text)

Signature

transcribe(audio: AudioInput, options?: SttOptions): Promise<Transcription>

Building an AudioInput

The constructors are exported as AudioInputs, plural, unlike every other SDK where the name is singular. Copied examples will not resolve.

AudioInputs.pcm16(data, sampleRate?, channels?)
AudioInputs.float32(data, sampleRate?, channels?)
AudioInputs.wav(data: Uint8Array, sampleRate = 16000)
AudioInputs.file(path: string, sampleRate = 16000)

AudioInputs.file is the common path: point it at a recording on disk.

With options

const transcription = await RunAnywhere.stt.transcribe(AudioInputs.file(path), {
  language: 'en',
  punctuation: true,
  wordTimestamps: true,
})

See STT options.

Checking readiness

const state = await RunAnywhere.stt.state()

Permission

Recording needs the microphone permission at runtime on both platforms. Request it before calling into the SDK; a denied permission surfaces as microphonePermissionDenied.