Natural, low-latency speech recognition for Capacitor apps with parity across iOS and Android, streaming partial results, and permission helpers baked in.

This package starts from the excellent capacitor-community/speech-recognition plugin, but folds in the most requested pull requests from that repo (punctuation support, segmented sessions, crash fixes) and keeps them maintained under the Capgo umbrella. You get the familiar API plus:

• ✅ Merged community PRs – punctuation toggles on iOS (PR #74), segmented results & silence handling on Android (PR #104), and the recognitionRequest safety fix (PR #105) ship out-of-the-box.
• 🚀 New Capgo features – configurable silence windows, streaming segment listeners, consistent permission helpers, and a refreshed example app.
• 🛠️ Active maintenance – same conventions as all Capgo plugins (SPM, Podspec, workflows, example app) so it tracks Capacitor major versions without bit-rot.
• 📦 Drop-in migration – TypeScript definitions remain compatible with the community plugin while exposing the extra options (addPunctuation, allowForSilence, segmentResults, etc.).

The most complete doc is available here: https://capgo.app/docs/plugins/speech-recognition/

Note: The major version of this plugin follows the major version of Capacitor. Use the version that matches your Capacitor installation (e.g., plugin v8 for Capacitor 8). Only the latest major version is actively maintained.

bun add @capgo/capacitor-speech-recognition
bunx cap sync

import { SpeechRecognition } from '@capgo/capacitor-speech-recognition';

await SpeechRecognition.requestPermissions();

const { available } = await SpeechRecognition.available();
if (!available) {
  console.warn('Speech recognition is not supported on this device.');
}

const partialListener = await SpeechRecognition.addListener('partialResults', (event) => {
  console.log('Partial:', event.matches?.[0]);
});

await SpeechRecognition.start({
  language: 'en-US',
  maxResults: 3,
  partialResults: true,
});

// Later, when you want to stop listening
await SpeechRecognition.stop();
await partialListener.remove();

This plugin now supports an opt-in on-device recognition path behind the explicit useOnDeviceRecognition flag.

The default path keeps the long-standing recognizer flow for backward compatibility. useOnDeviceRecognition switches to a newer local speech pipeline when the platform supports it:

• On iOS 26+, it uses Apple's SpeechAnalyzer / SpeechTranscriber stack.
• On recent Android versions, it uses the on-device SpeechRecognizer path.

• Better alignment with the latest native speech APIs.
• Improved on-device model handling on supported platforms.
• A cleaner rollout path if you want to adopt newer speech stacks without changing every user immediately.

Even when a new stack is technically available, changing recognition behavior silently can affect:

• transcript wording
• punctuation behavior
• partial-result timing
• product metrics and user expectations

That is why the plugin keeps the legacy recognizer by default and requires an explicit flag for the new path.

1. Check generic speech support with available().
2. Check the on-device path with isOnDeviceRecognitionAvailable().
3. Enable useOnDeviceRecognition only when that second check returns true.
4. Roll it out gradually if your app depends on stable transcripts or analytics.

import { SpeechRecognition } from '@capgo/capacitor-speech-recognition';

await SpeechRecognition.requestPermissions();

const { available } = await SpeechRecognition.available();
if (!available) {
  throw new Error('Speech recognition is not available on this device.');
}

const { available: onDeviceRecognitionAvailable } =
  await SpeechRecognition.isOnDeviceRecognitionAvailable({
    language: 'en-US',
  });

await SpeechRecognition.start({
  language: 'en-US',
  partialResults: true,
  useOnDeviceRecognition: onDeviceRecognitionAvailable,
});

Stay on the default path if:

• you need unchanged behavior for existing users
• you have not validated transcripts for your target locale
• you want identical production behavior across older and newer OS versions

• iOS uses the newer on-device path only on iOS 26+ and only for locales Apple exposes through the newer speech stack.
• Android uses the on-device recognizer only in inline mode. popup: true keeps using the system dialog and is not compatible with useOnDeviceRecognition.
• On Android, a supported on-device language may require a model download before recognition can begin.

This plugin also supports a push-to-talk oriented flow built around three APIs:

• setPTTState({ held }) lets your UI tell the plugin when the button is pressed or released.
• forceStop() stops the active session immediately and emits the last cached partial result with forced: true when available.
• getLastPartialResult() lets you read back the latest cached transcript at any point.

continuousPTT is the experimental cross-platform mode that keeps a held push-to-talk session alive by restarting recognition as speech segments finalize. Android and iOS both support this restart flow for inline/native recognition.

The plugin also emits deterministic session lifecycle events so UIs can react cleanly:

• listeningState now carries state, sessionId, reason, and optional errorCode in addition to the legacy status.
• error is emitted for every native recognizer error instead of relying only on promise rejections.
• readyForNextSession signals when native resources are torn down and the plugin is ready for another start.

Add the following keys to your app Info.plist:

• NSSpeechRecognitionUsageDescription
• NSMicrophoneUsageDescription

• available()
• isOnDeviceRecognitionAvailable(...)
• start(...)
• stop()
• forceStop(...)
• getLastPartialResult()
• setPTTState(...)
• getSupportedLanguages()
• isListening()
• checkPermissions()
• requestPermissions()
• getPluginVersion()
• addListener('endOfSegmentedSession', ...)
• addListener('segmentResults', ...)
• addListener('partialResults', ...)
• addListener('listeningState', ...)
• addListener('error', ...)
• addListener('readyForNextSession', ...)
• removeAllListeners()
• Interfaces
• Type Aliases

available() => Promise<SpeechRecognitionAvailability>

Checks whether the native speech recognition service is usable on the current device.

Returns: Promise<SpeechRecognitionAvailability>

isOnDeviceRecognitionAvailable(options?: Pick<SpeechRecognitionStartOptions, "language"> | undefined) => Promise<SpeechRecognitionAvailability>

Checks whether the platform's newer on-device recognition path is available for the selected locale.

This is the capability check you should use before enabling useOnDeviceRecognition. A true result means the current device, OS version, and locale can use the newer on-device path for that platform.

Returns false when the device only supports the legacy recognizer path.

Platform SDK docs: iOS: Speech Android: SpeechRecognizer

Returns: Promise<SpeechRecognitionAvailability>

start(options?: SpeechRecognitionStartOptions | undefined) => Promise<SpeechRecognitionMatches>

Begins capturing audio and transcribing speech.

When partialResults is true, the returned promise resolves immediately and updates are streamed through the partialResults listener until the session ends.

The default path keeps the legacy recognizer behavior for backward compatibility. Pass useOnDeviceRecognition: true only after checking {@link SpeechRecognitionPlugin.isOnDeviceRecognitionAvailable}.

Returns: Promise<SpeechRecognitionMatches>

stop() => Promise<void>

Stops listening and tears down native resources.

forceStop(options?: ForceStopOptions | undefined) => Promise<void>

Force stops the current session.

On Android, this first tries a normal stop and then falls back to destroy/recreate after timeout. On iOS, the current session is stopped immediately.

If a partial transcript is cached, it is emitted through the partialResults listener with forced: true.

getLastPartialResult() => Promise<LastPartialResult>

Gets the last cached partial transcription result.

Returns: Promise<LastPartialResult>

setPTTState(options: PTTStateOptions) => Promise<void>

Updates the current push-to-talk button state.

Use this together with continuousPTT or with a custom hold-to-talk flow.

getSupportedLanguages() => Promise<SpeechRecognitionLanguages>

Gets the locales supported by the underlying recognizer.

Android 13+ devices no longer expose this list; in that case languages is empty.

Returns: Promise<SpeechRecognitionLanguages>

isListening() => Promise<SpeechRecognitionListening>

Returns whether the plugin is actively listening for speech.

Returns: Promise<SpeechRecognitionListening>

checkPermissions() => Promise<SpeechRecognitionPermissionStatus>

Gets the current permission state.

Returns: Promise<SpeechRecognitionPermissionStatus>

requestPermissions() => Promise<SpeechRecognitionPermissionStatus>

Requests the microphone + speech recognition permissions.

Returns: Promise<SpeechRecognitionPermissionStatus>

getPluginVersion() => Promise<{ version: string; }>

Returns the native plugin version bundled with this package.

Useful when reporting issues to confirm that native and JS versions match.

Returns: Promise<{ version: string; }>

addListener(eventName: 'endOfSegmentedSession', listenerFunc: () => void) => Promise<PluginListenerHandle>

Listen for segmented session completion events (Android only).

Returns: Promise<PluginListenerHandle>

addListener(eventName: 'segmentResults', listenerFunc: (event: SpeechRecognitionSegmentResultEvent) => void) => Promise<PluginListenerHandle>

Listen for segmented recognition results (Android only).

Returns: Promise<PluginListenerHandle>

addListener(eventName: 'partialResults', listenerFunc: (event: SpeechRecognitionPartialResultEvent) => void) => Promise<PluginListenerHandle>

Listen for partial transcription updates emitted while partialResults is enabled.

Returns: Promise<PluginListenerHandle>

addListener(eventName: 'listeningState', listenerFunc: (event: SpeechRecognitionListeningEvent) => void) => Promise<PluginListenerHandle>

Listen for changes to the native listening state.

Returns: Promise<PluginListenerHandle>

addListener(eventName: 'error', listenerFunc: (event: SpeechRecognitionErrorEvent) => void) => Promise<PluginListenerHandle>

Listen for recognition errors.

Returns: Promise<PluginListenerHandle>

addListener(eventName: 'readyForNextSession', listenerFunc: (event: SpeechRecognitionReadyEvent) => void) => Promise<PluginListenerHandle>

Listen for the recognizer becoming ready for another session.

Returns: Promise<PluginListenerHandle>

removeAllListeners() => Promise<void>

Removes every registered listener.

Configure how the recognizer behaves when calling {@link SpeechRecognitionPlugin.start}.

Options for {@link SpeechRecognitionPlugin.forceStop}.

Result from {@link SpeechRecognitionPlugin.getLastPartialResult}.

Options for {@link SpeechRecognitionPlugin.setPTTState}.

Permission map returned by checkPermissions and requestPermissions.

On Android the state maps to the RECORD_AUDIO permission. On iOS it combines speech recognition plus microphone permission.

Raised whenever a segmented result is produced (Android only).

Raised whenever a partial transcription is produced.

Raised when the listening state changes.

The original status field is preserved for backward compatibility and is present on the binary started / stopped states.

Raised whenever native recognition reports an error.

Emitted after native resources have been torn down and the plugin is ready for another session.

From T, pick a set of properties whose keys are in the union K

{ [P in K]: T[P]; }

'prompt' | 'prompt-with-rationale' | 'granted' | 'denied'

Finite state values for the recognition session lifecycle.

'startingListening' | 'started' | 'stoppingListening' | 'stopped'

Why a listening state transition happened.

'userStart' | 'userStop' | 'forceStop' | 'results' | 'silence' | 'error' | 'unknown'