flutter_native_speech_to_text 1.3.0

Speech-to-text for Flutter using Android RecognizerIntent, Apple Speech on iOS and macOS, the Windows recognizer, Vosk on Linux and the Web Speech API.

flutter_native_speech_to_text

A lightweight Flutter plugin for speech-to-text recognition with one Dart API on every platform: Android's native RecognizerIntent.ACTION_RECOGNIZE_SPEECH, Apple's Speech framework on iOS and macOS, the Windows speech recognizer, Vosk on Linux, and the browser's Web Speech API on web.

---

Features

• 🎯 Native Android Integration: Direct access to Android's built-in RecognizerIntent.ACTION_RECOGNIZE_SPEECH dialog.
• 🍎 iOS and macOS: Uses Apple's Speech framework, with no extra dependencies.
• 🌍 Web Support: Runs in Chrome, Edge and Safari through the Web Speech API with the same Dart API.
• 🔎 Language Check: isLanguageSupported() tells you whether a language works on the current platform.
• 🖥️ Desktop: Windows uses the built-in recognizer; Linux uses Vosk when installed.
• ⚙️ Fully Configurable: Exposes all major RecognizerIntent extras as typed Dart parameters.
• 🌐 Multi-language Support: Specify custom language codes (e.g., en-US, bn-BD, es-ES).
• 📊 Detailed Results: Access primary transcription, alternate transcript hypotheses, and confidence scores.
• ⏱️ Silence & Length Controls: Fine-tune speech input minimum length and silence thresholds.
• 🪶 Zero Bloat: Lightweight implementation using each platform's own speech services, without heavy third-party models.

---

Platform Support

Platform	Supported	Notes
Android	✅	Requires device with Google Play Services or Speech Recognizer installed (API 24+)
iOS	✅	Apple Speech framework; Bengali is not supported by Apple (see iOS and macOS)
macOS	✅	Apple Speech framework (see iOS and macOS)
Web	✅	Uses the browser's Web Speech API (see Web behaviour)
Windows	⚠️	Windows speech recognizer (untested, see Windows)
Linux	⚠️	Vosk + PulseAudio, loaded at runtime (untested, see Linux)

---

Installation

Add flutter_native_speech_to_text to your pubspec.yaml:

dependencies:
  flutter_native_speech_to_text: ^1.3.0

Or run:

flutter pub add flutter_native_speech_to_text

---

Android Setup

1. Permissions & Queries

Add the following permissions and <queries> declaration to your android/app/src/main/AndroidManifest.xml (inside the <manifest> tag):

<manifest xmlns:android="http://schemas.android.com/apk/res/android">

    <!-- Permissions required for speech recognition -->
    <uses-permission android:name="android.permission.RECORD_AUDIO" />
    <uses-permission android:name="android.permission.INTERNET" />

    <!-- Queries declaration for Android 11+ (API 30+) package visibility -->
    <queries>
        <intent>
            <action android:name="android.speech.action.RECOGNIZE_SPEECH" />
        </intent>
    </queries>

    <application ...>
        ...
    </application>
</manifest>

Note on Android 6.0+ (API 23+): Make sure your application requests the runtime RECORD_AUDIO permission before initiating speech recognition if needed, or handle permission callbacks accordingly.

---

Usage

1. Check Availability

Before starting recognition, check if a speech recognizer service is available on the user's device:

import 'package:flutter_native_speech_to_text/flutter_native_speech_to_text.dart';

final bool isAvailable = await FlutterNativeSpeechToText.isAvailable();

if (!isAvailable) {
  print('Speech recognition is not available on this device.');
}

2. Check Language Support (optional)

final bool ok = await FlutterNativeSpeechToText.isLanguageSupported('en-US');

See Checking language support for how each platform answers.

3. Start Speech Recognition

Invoke FlutterNativeSpeechToText.listen(). Android opens the system speech dialog; iOS, macOS and web record without showing a dialog. Pass language explicitly on iOS, macOS and web if you don't use the default bn-BD:

try {
  final SpeechResult result = await FlutterNativeSpeechToText.listen(
    language: 'en-US',
    languageModel: 'free_form', // 'free_form' or 'web_search'
    prompt: 'Say something...',
    maxResults: 5,
    partialResults: true,
    confidenceScores: true,
    speechInputMinimumLengthMillis: 2000,
    speechInputCompleteSilenceLengthMillis: 1500,
    speechInputPossiblyCompleteSilenceLengthMillis: 500,
  );

  print('Top Result: ${result.recognizedText}');
  print('Alternatives: ${result.alternativeResults}');
  print('Confidence Scores: ${result.confidenceScores}');
} on SpeechToTextException catch (e) {
  print('Error code: ${e.code}, message: ${e.message}');
}

4. Stop or Cancel Recognition

// Stop listening
await FlutterNativeSpeechToText.stop();

// Cancel listening session
await FlutterNativeSpeechToText.cancel();

---

API & Parameter Reference

FlutterNativeSpeechToText.listen() Parameters

Parameter	RecognizerIntent Equivalent	Type	Description
language	EXTRA_LANGUAGE	String?	Recognition language tag (e.g. 'en-US', 'bn-BD'). Defaults to 'bn-BD'.
languagePreference	EXTRA_LANGUAGE_PREFERENCE	String?	Preferred language for recognition.
languageModel	EXTRA_LANGUAGE_MODEL	String?	'free_form' (default) or 'web_search'.
prompt	EXTRA_PROMPT	String?	Prompt text displayed to the user in the system dialog.
maxResults	EXTRA_MAX_RESULTS	int?	Maximum number of alternative recognition results.
partialResults	EXTRA_PARTIAL_RESULTS	bool?	Request partial recognition results.
confidenceScores	EXTRA_CONFIDENCE_SCORES	bool?	Request confidence scores for transcriptions.
speechInputMinimumLengthMillis	EXTRA_SPEECH_INPUT_MINIMUM_LENGTH_MILLIS	int?	Minimum duration in milliseconds to consider speech input.
speechInputCompleteSilenceLengthMillis	EXTRA_SPEECH_INPUT_COMPLETE_SILENCE_LENGTH_MILLIS	int?	Silence duration to consider speech complete.
speechInputPossiblyCompleteSilenceLengthMillis	EXTRA_SPEECH_INPUT_POSSIBLY_COMPLETE_SILENCE_LENGTH_MILLIS	int?	Silence duration to consider speech possibly complete.
getAudio	EXTRA_GET_AUDIO	bool?	Whether to request recorded audio data.
getAudioFormat	EXTRA_GET_AUDIO_FORMAT	String?	MIME type format for requested audio (e.g., 'audio/AMR').
originatorPackage	EXTRA_ORIGINATOR_PKG	String?	Calling package name.
isEnable	N/A	bool	Set to false to disable recognition (throws error). Defaults to true.

Note: Unspecified parameters use each platform's defaults. Parameters a platform does not support are ignored (see the web, iOS and macOS sections).

---

Checking language support

Use isLanguageSupported before listen() to find out whether a language works on the current platform:

if (await FlutterNativeSpeechToText.isLanguageSupported('bn-BD')) {
  final result = await FlutterNativeSpeechToText.listen(language: 'bn-BD');
} else {
  // Fall back to another language or show a message.
}

Platform	Result
iOS / macOS	true only for languages in Apple's supported list (bn-BD is false)
Android	Uses the recognizer's language list. If the device reports none, returns true
Web	true when the browser has speech recognition. Browsers do not list languages, so an unsupported language is reported by listen() instead
Windows	true for languages in the Windows speech recognizer's list
Linux	true when a model folder for the language exists in VOSK_MODEL_PATH

A bare code such as en matches any region (en-US, en-GB, ...).

---

Result Model (SpeechResult)

Property	Type	Description
recognizedText	String?	The most probable recognized text string.
alternativeResults	List<String>?	List of alternative transcription hypotheses.
confidenceScores	List<double>?	Confidence scores corresponding to alternative results (0.0 to 1.0).
isFinal	bool	Indicates whether the result is final.

---

Error Codes

When a SpeechToTextException is thrown, the code property provides specific failure reasons:

Code	Reason
NO_RECOGNIZER	No speech recognizer service is installed or active on the device (on Linux, Vosk or PulseAudio is missing).
NO_CONTEXT	Android activity context was null or detached.
ALREADY_LISTENING	A speech recognition session is already in progress.
DISABLED	The isEnable flag was passed as false.
ERROR	An exception occurred while launching the intent.
PLATFORM_NOT_SUPPORTED	The platform has no implementation.
PERMISSION_DENIED	(Web) Microphone permission was denied.
SERVICE_NOT_ALLOWED	(Web) The browser blocked its speech service, for example Safari with Dictation off or the language not installed.
NO_MICROPHONE	(Web) No usable microphone was found.
NETWORK	(Web) The browser's speech service could not be reached.
LANGUAGE_NOT_SUPPORTED	(Web, iOS, macOS) The requested language is not supported by the engine.
MISSING_PLUGIN	The native side of the plugin is not registered in the current binary (e.g. plugin missing from a release build).

---

Web behaviour

On web the plugin uses the browser's Web Speech API. No setup or extra dependencies are needed - the same listen(), stop(), cancel() and isAvailable() calls work unchanged.

• Browser support:
  • Chrome and Edge: supported, including languages such as bn-BD.
  • Safari: supported, but it uses the macOS speech engine, so Dictation must be turned on and the requested language must be installed under System Settings

    Keyboard > Dictation. Otherwise listen() throws SERVICE_NOT_ALLOWED.

  • Brave: blocks the speech service, so isAvailable() returns false and listen() throws NO_RECOGNIZER.
  • Firefox: does not implement speech recognition, so isAvailable() returns false and listen() throws NO_RECOGNIZER.
• Secure context: the page must be served over HTTPS (or localhost), and the user must grant microphone permission when prompted.
• Server-side recognition: Chrome sends audio to Google's servers, so an internet connection is required.
• Single utterance: like Android, listen() completes after one utterance (or when you call stop()); cancel() completes it with an empty result.
• Supported parameters: language / languagePreference (default bn-BD), partialResults (enables interim hypotheses internally), maxResults (alternatives count) and isEnable. Android-only options (prompt, silence lengths, getAudio, ...) are ignored.
• confidenceScores and alternativeResults are filled from the browser's result.

---

iOS and macOS

On iOS and macOS the plugin uses Apple's Speech framework (SFSpeechRecognizer) and the microphone. No dialog is shown: listen() starts recording, stops after about two seconds of silence (speechInputCompleteSilenceLengthMillis), and returns the result. stop() ends the session early and cancel() completes it with an empty result.

iOS (ios/Runner/Info.plist):

<key>NSMicrophoneUsageDescription</key>
<string>Used to recognise your speech.</string>
<key>NSSpeechRecognitionUsageDescription</key>
<string>Used to convert your speech to text.</string>

macOS: add the same two keys to macos/Runner/Info.plist, and these entitlements to DebugProfile.entitlements and Release.entitlements:

<key>com.apple.security.device.audio-input</key>
<true/>
<key>com.apple.security.network.client</key>
<true/>

Notes:

• Languages: only the languages Apple supports are available. Bengali (bn-BD) is not supported by Apple's speech engine, so listen() throws LANGUAGE_NOT_SUPPORTED unless you pass another language, for example en-US. The default language is bn-BD, so pass language explicitly on Apple platforms.
• Supported parameters: language / languagePreference, maxResults, speechInputCompleteSilenceLengthMillis and isEnable. Other Android-only options are ignored.
• Some languages need an internet connection for recognition.

Windows

Uses the built-in Windows speech recognizer (Windows.Media.SpeechRecognition). No dialog is shown. listen() records until about five seconds of silence, or the value of speechInputCompleteSilenceLengthMillis, then returns the result.

• Turn on Settings > Privacy & security > Speech > Online speech recognition, otherwise listen() throws PERMISSION_DENIED.
• Allow desktop apps to use the microphone in Settings > Privacy & security > Microphone.
• Only the languages installed for Windows speech recognition work. Other languages, which may include Bengali, throw LANGUAGE_NOT_SUPPORTED.
• Supported parameters: language / languagePreference, maxResults, speechInputCompleteSilenceLengthMillis and isEnable.
• This implementation has not been tested on a Windows machine yet.

Linux

Linux has no built-in speech engine, so the plugin uses Vosk for recognition and PulseAudio for the microphone. Both are loaded at runtime, so your app builds and runs without them and only needs them for recognition:

1. Install libvosk.so (so the loader can find it) and libpulse-simple (for example the libpulse0 package).
2. Download Vosk models and put each in its own folder named after its language tag under one directory, for example models/en-us and models/bn-bd.
3. Set the VOSK_MODEL_PATH environment variable to that directory.

Without these, isAvailable() returns false and listen() throws NO_RECOGNIZER. A language with no model folder throws LANGUAGE_NOT_SUPPORTED. Recognition ends when Vosk detects the end of an utterance, so speechInputCompleteSilenceLengthMillis is ignored, and confidenceScores is not filled. Vosk's available languages depend on the models you download. This implementation has not been tested on Linux yet.

---

Example

For a complete working sample application, check out the example folder in this repository.

---

License

This project is licensed under the MIT License - see the LICENSE file for details.