whisper_flutter_coreml 1.0.1

A flutter library for offline speech-to-text conversion which use whisper.cpp models implementation for Android、iOS、macOS.

Whisper Flutter CoreML

Ready to use whisper.cpp models implementation for iOS and Android

1. Support AGP8+
2. Support Android 5.0+ & iOS 13+ & MacOS 11+
3. It is optimized and fast

Supported models: tiny、base、small、medium、large-v1、large-v2, large-v3-turbo

Recommended Models：base、small、medium, turbo

All models have been actually tested, test devices: Android: Google Pixel 7 Pro, iOS: M1 iOS simulator，MacOS: M1 MacBookPro & M2 MacMini

Install library

flutter pub add whisper_flutter_coreml

import library

import 'package:whisper_flutter_coreml/whisper_flutter_coreml.dart';

CoreML Hardware Acceleration

Whisper Flutter CoreML now supports CoreML hardware acceleration on iOS 13+ and macOS 11+ devices with Apple Silicon. This enables significantly faster inference by utilizing the Apple Neural Engine (ANE).

Performance Benefits

• Up to 3x faster transcription on Apple Silicon devices
• Lower power consumption compared to CPU-only execution
• Optimized for Apple Neural Engine on supported devices

Requirements for CoreML

• iOS 13.0+ or macOS 11.0+
• Apple Silicon device (M1, M2, M3, or newer recommended)
• CoreML model files (.mlmodelc format)

CoreML Model Generation

To use CoreML acceleration, you'll need to generate CoreML models from your Whisper models. The models will automatically fallback to CPU execution if CoreML models are not available.

Note: CoreML model generation requires a separate setup process. The first run with CoreML may be slower as the Apple Neural Engine compiles the model for your specific device.

Quickstart

// Prepare wav file
final Directory documentDirectory = await getApplicationDocumentsDirectory();
final ByteData documentBytes = await rootBundle.load('assets/jfk.wav');

final String jfkPath = '${documentDirectory.path}/jfk.wav';

await File(jfkPath).writeAsBytes(
    documentBytes.buffer.asUint8List(),
);

// Begin whisper transcription
/// China: https://hf-mirror.com/ggerganov/whisper.cpp/resolve/main
/// Other: https://huggingface.co/ggerganov/whisper.cpp/resolve/main
final Whisper whisper = Whisper(
    model: WhisperModel.base,
    downloadHost: "https://huggingface.co/ggerganov/whisper.cpp/resolve/main"
);

final String? whisperVersion = await whisper.getVersion();
print(whisperVersion);

final String transcription = await whisper.transcribe(
    transcribeRequest: TranscribeRequest(
        audio: jfkPath,
        isTranslate: true, // Translate result from audio lang to english text
        isNoTimestamps: false, // Get segments in result
        splitOnWord: true, // Split segments on each word 
    ),
);
print(transcription);