flutter_whisper_kit #

A Flutter plugin that provides on-device speech recognition capabilities using WhisperKit.

Features #

• High-quality on-device speech recognition
• No data sent to external servers
• Support for multiple model sizes (tiny to large)
• File-based audio transcription
• Real-time microphone transcription
• Progress tracking for model downloads
• Configurable transcription options

Demo #

https://github.com/user-attachments/assets/4a405460-2eb9-4485-a467-24b8ebf6bee7

Installation #

Add the following to your pubspec.yaml:

dependencies:
  flutter_whisper_kit: latest

Usage #

Basic Usage #

import 'package:flutter_whisper_kit/flutter_whisper_kit.dart';

// Create an instance of the plugin
final whisperKit = FlutterWhisperKit();

// Load a model
final result = await whisperKit.loadModel(
  variant: 'tiny-en',
  modelRepo: 'argmaxinc/whisperkit-coreml',
  redownload: false,
);
print('Model loaded: $result');

// Transcribe from a file
final transcription = await whisperKit.transcribeFromFile(
  '/path/to/audio/file.mp3',
  options: DecodingOptions(
    task: DecodingTask.transcribe,
    language: 'en',
  ),
);
print('Transcription: $transcription');

Real-time Transcription #

import 'package:flutter_whisper_kit/flutter_whisper_kit.dart';

// Create an instance of the plugin
final whisperKit = FlutterWhisperKit();

// Load a model
await whisperKit.loadModel(variant: 'tiny-en');

// Start listening to the transcription stream
whisperKit.transcriptionStream.listen((transcription) {
  print('Real-time transcription: $transcription');
});

// Start recording
await whisperKit.startRecording(
  options: DecodingOptions(
    task: DecodingTask.transcribe,
    language: 'en',
  ),
);

// Stop recording after some time
await Future.delayed(Duration(seconds: 10));
final finalTranscription = await whisperKit.stopRecording();
print('Final transcription: $finalTranscription');

Platform Support #

Currently, Flutter WhisperKit supports:

• iOS 16.0+
• macOS 13.0+

Android support is planned for future releases.

License #

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