@xsai-apple-speech/transcription

This package owns the platform-neutral Apple Speech Provider contract, xsAI batch adapter, locale rules, structured errors, and live Web Streams lifecycle.

Use it when you consume a Provider or implement a new native or remote transport. Use @xsai-apple-speech/transcription-native for direct macOS execution. Use @xsai-apple-speech/transcription-electron-plugin in a renderer.

import { streamTranscription } from '@xsai-apple-speech/transcription'

const session = streamTranscription({
  ...provider.transcription({ locale: 'zh-CN' }),
  inputSampleRate: 48_000,
})

partialStream yields replacement text for the current partial result. fullStream yields ordered partial events and one final transcript.text.done event. done and text settle only after the implementation releases its session resources.

The transcriber option defaults to automatic. Automatic mode prefers SpeechTranscriber and falls back to DictationTranscriber for an unsupported locale.

The options object groups text, reporting, attribute, and content-hint overrides. Automatic mode accepts the union of both transcriber option sets.

const transcription = provider.transcription({
  locale: 'en-US',
  options: {
    transcription: {
      includePunctuation: true,
    },
  },
})

Native result chunks are available in TranscriptionResult.results. A partial event exposes its changed native chunk through result.

Apple Speech adds alternatives and attributed metadata only when their options are enabled. The main text field stays compatible with xsAI transcription calls.

Use createAbortError(message) for cancellation that crosses Node.js, Electron, or Eventa boundaries. It returns a serializable Error whose name is AbortError.

Do not pass a browser MediaStream to this package. Convert captured audio to mono Float32Array PCM in the application.