Platform-Conditional STT Provider

Use this pattern when targeting both iOS and Android but want platform-optimized transcription: Apple's on-device recognition on iOS, whisper.rn offline inference on Android.

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

When to Use

• You want best-in-class iOS accuracy without shipping a 75 MB model on iOS
• You still need fully offline transcription on Android
• You are comfortable that iOS results come from Apple's speech servers (network-dependent)

Pattern

import { Platform } from 'react-native';
import { initialize } from 'react-native-voice-activator';
import { WhisperRNSTTAdapter } from 'react-native-voice-activator/whisper-rn';

// iOS path — app-owned adapter wrapping expo-speech-recognition
// See: docs/examples/expo-speech-recognition-stt-provider.md
import { ExpoSpeechRecognitionSttProvider } from './ExpoSpeechRecognitionSttProvider';

// Android path — on-device whisper.rn
// See: docs/examples/whisper-stt-provider.md
const whisperAdapter = new WhisperRNSTTAdapter({
  modelUrl: 'https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin',
  language: 'en',
});

await initialize({
  sttProvider:
    Platform.OS === 'ios'
      ? new ExpoSpeechRecognitionSttProvider(bridge)
      : whisperAdapter,
});

iOS: Required Permission

When using expo-speech-recognition (or any SFSpeechRecognizer-based adapter) on iOS you must add NSSpeechRecognitionUsageDescription to your Info.plist in addition to the standard microphone permission:

Bare React Native — ios/<App>/Info.plist:

<key>NSSpeechRecognitionUsageDescription</key>
<string>This app uses speech recognition to transcribe what you say.</string>
<key>NSMicrophoneUsageDescription</key>
<string>This app uses the microphone to detect wake words.</string>

Expo — app.json / app.config.js:

{
  "expo": {
    "ios": {
      "infoPlist": {
        "NSSpeechRecognitionUsageDescription": "This app uses speech recognition to transcribe what you say."
      }
    }
  }
}

Tradeoffs

	iOS (expo-speech-recognition)	Android (whisper.rn)
Accuracy	Apple models, high accuracy	Good; depends on model size
Latency	Fast (Apple Neural Engine)	Slower; CPU-bound on older devices
Network	Required (cloud-backed by default)	Not required (fully offline)
Model download	None	~75 MB (tiny.en) to ~1.5 GB (large)
Extra permission	NSSpeechRecognitionUsageDescription	None beyond microphone
Language support	Apple's supported locales	100+ via Whisper model choice

Real Constraints

• iOS recognition requires network; the adapter should handle the offline fallback if your app must work without connectivity on iOS
• expo-speech-recognition handles Android too (Google Speech), but that path is also cloud-backed — the pattern above deliberately uses whisper.rn on Android for offline support
• vendor setup, permission requests, and locale configuration are app-owned concerns; see the individual adapter docs for each platform path