Expo Speech Recognition STT Provider

This reference adapter shows how an Expo-oriented app can wrap an expo-speech-recognition style integration behind the public SpeechToTextProvider interface without moving vendor wiring into the package.

Use this in the host app only. Do not move it into the library package core.

Integration Notes

• keep the speech-recognition library installed in the consuming app, not in react-native-voice-activator
• pass the adapter through initialize({ sttProvider })
• build a small app-owned bridge around the exact vendor version you install
• recognition setup, permissions, locale handling, and network/offline behavior remain vendor- and app-specific concerns

Reference Adapter

import type {
  SpeechToTextProvider,
  TranscriptionResult,
} from 'react-native-voice-activator';

type ExpoSpeechRecognitionBridge = {
  startSingleUtterance(): Promise<{
    transcript: string;
    confidence?: number;
    durationMs?: number;
  }>;
  cancel(): Promise<void>;
};

export class ExpoSpeechRecognitionSttProvider
  implements SpeechToTextProvider
{
  readonly name = 'expo-speech-recognition';

  constructor(private readonly bridge: ExpoSpeechRecognitionBridge) {}

  async transcribe(): Promise<TranscriptionResult> {
    const result = await this.bridge.startSingleUtterance();

    return {
      text: result.transcript,
      confidence: result.confidence,
      durationMs: result.durationMs,
      provider: this.name,
    };
  }

  async cancel(): Promise<void> {
    await this.bridge.cancel();
  }
}

Implement the bridge in your app against the exact vendor version you ship. For example, your bridge can own:

• permission requests
• listener registration and cleanup
• the package-specific start/stop calls
• conversion from vendor result events into a single resolved transcript

Usage

import { initialize } from 'react-native-voice-activator';

await initialize({
  sttProvider: new ExpoSpeechRecognitionSttProvider(),
});

Real Constraints

• recognition availability differs across platforms and OS versions
• cloud-backed recognition is not equivalent to on-device offline STT
• the host app must own microphone permission flow and any vendor-specific setup
• this repo does not validate the vendor package API for you; it only locks the adapter boundary shown above
• if your app needs an offline path, prefer an on-device adapter such as the architecture-approved Sherpa or ONNX-based options in app-owned code