@honeypathkar/react-native-speech-to-text

Native speech-to-text for React Native on iOS and Android, with configurable silence detection that commits the transcript and stops the microphone once the user has paused.

• Real-time interim transcripts (SFSpeechRecognizer on iOS, SpeechRecognizer on Android)
• Native silence / pause detection with auto-stop (default 2.5s)
• Audio level metering for waveforms and mic meters
• Every recognition locale the device supports, with display names
• Optional offline / on-device recognition
• Promise API, typed event stream, and a useSpeechToText hook

Install

npm install @honeypathkar/react-native-speech-to-text
cd ios && pod install

Android is picked up by autolinking; just rebuild.

iOS setup

Add both keys to ios/<YourApp>/Info.plist:

<key>NSSpeechRecognitionUsageDescription</key>
<string>Used to transcribe what you say into text.</string>
<key>NSMicrophoneUsageDescription</key>
<string>Used to record your voice for transcription.</string>

Minimum deployment target: iOS 13.0.

Android setup

RECORD_AUDIO and the <queries> entry for RecognitionService ship in the library manifest and merge into your app automatically — no manual edits needed. requestPermissions() handles the runtime prompt.

Minimum SDK: 21. On-device recognition needs API 31+.

Quick start

import SpeechToText from '@honeypathkar/react-native-speech-to-text';

const { granted } = await SpeechToText.requestPermissions();
if (!granted) return;

SpeechToText.on('onSpeechPartialResults', ({ transcript }) => {
  console.log('Interim:', transcript);
});

SpeechToText.on('onSpeechResults', ({ transcript }) => {
  console.log('Final:', transcript);
});

SpeechToText.on('onSpeechEnd', ({ reason }) => {
  console.log('Stopped because:', reason); // 'silence' | 'manual' | ...
});

await SpeechToText.startListening({
  locale: 'en-US',
  silenceTimeoutMs: 2500, // auto-stop after a 2.5s pause
  interimResults: true,
});

stopListening() also resolves with the final transcript, if you would rather await it than listen for the event:

const { transcript, reason } = await SpeechToText.stopListening();

Hook

import { useSpeechToText } from '@honeypathkar/react-native-speech-to-text';

function VoiceInput() {
  const {
    isListening,
    transcript,
    interimTranscript,
    audioLevel,
    error,
    startListening,
    stopListening,
  } = useSpeechToText({
    locale: 'en-US',
    silenceTimeoutMs: 2500,
    onResult: (text) => console.log('committed:', text),
  });

  return (
    <View>
      <Text>{transcript || interimTranscript || 'Say something…'}</Text>
      <View style={{ width: 200 * audioLevel, height: 4, backgroundColor: '#0a0' }} />
      <Button
        title={isListening ? 'Stop' : 'Start'}
        onPress={isListening ? stopListening : () => startListening()}
      />
      {error ? <Text>{error.message}</Text> : null}
    </View>
  );
}

The hook removes its listeners and tears down any live session on unmount.

Silence detection

silenceDetectionMode controls what counts as "still talking":

| Mode | Timer resets on | Trade-off | | --- | --- | --- | | transcript (default) | New words from the recognizer | Immune to background noise | | audio | Input level above threshold | Most responsive; a noisy room can hold the mic open | | hybrid | Either signal | Least likely to cut off a slow speaker; stops latest |

transcript is the default because in a noisy room the audio level alone rarely drops below the threshold, so audio and hybrid can keep listening forever.

Android's EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS is advisory and widely ignored by OEM recognition services, so the pause is timed in the library on both platforms. That keeps silenceTimeoutMs meaning the same thing everywhere.

Set continuous: true to keep the microphone open through pauses — you still get an onSpeechSilence event on every pause, but the session only ends when you call stopListening().

API

Methods

| Method | Description | | --- | --- | | requestPermissions() | Prompts for microphone (and speech recognition on iOS). | | getPermissionStatus() | Reads permission state without prompting. | | isAvailable() | Whether recognition works right now. | | isAvailableForLocale(locale) | Whether one BCP-47 locale is usable. | | supportsOnDeviceRecognition(locale?) | Whether offline recognition is available. | | getAvailableLocales() | All supported locales with display names. | | getAvailableLanguages() | All supported locales as BCP-47 tags. | | startListening(options?) | Opens the mic and starts transcribing. | | stopListening() | Stops and resolves with the final transcript. | | cancel() | Stops and discards the transcript. | | destroy() | Tears down and releases the audio hardware. | | on(event, listener) | Subscribes; returns a subscription with .remove(). | | removeAllListeners(event?) | Drops listeners. |

startListening options

| Option | Default | Description | | --- | --- | --- | | locale | device locale | BCP-47 tag, e.g. 'en-US', 'hi-IN', 'gu-IN'. | | silenceTimeoutMs | 2500 | Pause length that triggers auto-stop. | | interimResults | true | Stream partial transcripts. | | continuous | false | Keep listening through pauses. | | requiresOnDeviceRecognition | false | Force offline. Rejects if unavailable. | | silenceDetectionMode | 'transcript' | transcript | audio | hybrid. | | silenceThresholdDb | -35 | iOS silence threshold, dBFS. | | androidSilenceThresholdRms | 2 | Android silence threshold, RMS scale. | | noSpeechTimeoutMs | 0 (off) | Give up if the user never speaks. | | maxDurationMs | 0 (off) | Hard cap on one session. | | addsPunctuation | true | iOS 16+ automatic punctuation. | | taskHint | 'unspecified' | iOS recognition hint. | | contextualStrings | – | Bias toward names/jargon. Android needs API 33+. | | volumeUpdates | true | Emit onSpeechVolumeChanged. | | volumeIntervalMs | 100 | Throttle for volume events. |

Events

| Event | Payload | | --- | --- | | onSpeechReady | { locale, silenceTimeoutMs, onDevice } | | onSpeechStart | { timestamp } — the user actually started speaking | | onSpeechPartialResults | { transcript, isFinal: false, segments } | | onSpeechResults | { transcript, isFinal: true } | | onSpeechSilence | { durationMs, transcript } | | onSpeechEnd | { reason, transcript } | | onSpeechError | { code, message, nativeCode? } | | onSpeechVolumeChanged | { value, db } — value is 0..1 |

reason is one of silence, manual, no_speech, max_duration, recognizer_final, cancelled, destroyed, error, not_listening.

Error codes

permission_denied, already_listening, recognizer_unavailable, locale_not_supported, on_device_unavailable, no_match, no_speech_detected, network_error, network_timeout, recognizer_busy, audio_error, recognition_error.

Platform differences

| | iOS | Android | | --- | --- | --- | | Engine | SFSpeechRecognizer + AVAudioEngine | SpeechRecognizer | | Word-level segments | Yes | Empty array | | Volume scale (db) | dBFS | onRmsChanged scale | | Speech permission | Separate from mic | Mirrors mic | | Offline | iOS 13+ where supported | API 31+ | | continuous | One session stays open | Recognizer restarts, transcript accumulates |

value on onSpeechVolumeChanged is normalized to 0..1 on both platforms, so prefer it over db for meters.

License

MIT