qwen_asr 0.2.6

On-device Qwen3-ASR speech-to-text for Flutter. Pure Rust, no cloud, no server.

qwen_asr WIP

Work in Progress — This library is still under active development. Use it at your own risk. APIs may change and some features may be incomplete or unstable. I will continue to improve it over time.

On-device Qwen3-ASR speech-to-text for Flutter. Runs entirely on the device using a pure Rust inference engine — no cloud services, no server, no network required. Supports iOS, Android, and macOS.

The native Rust engine is compiled in release mode automatically by the Flutter build pipeline. If building the Rust library manually, always use --release — debug builds are 10–50x slower and unusable for real-time inference.

Setup

Add the dependency:

dependencies:
  qwen_asr: ^0.1.0

Download the model (e.g. in your app's asset pipeline or at first launch):

huggingface_hub download Qwen/Qwen3-ASR-0.6B --local-dir <your-model-dir>

Usage

import 'package:qwen_asr/qwen_asr.dart';

// Load the model (once, typically at app start)
final engine = await QAsrEngine.load('/path/to/qwen3-asr-0.6b');

// Transcribe a WAV file
final text = await engine.transcribeFile('/path/to/audio.wav');
print(text);

// Transcribe raw PCM (Float32List, 16 kHz, mono, values in -1..1)
final pcmText = await engine.transcribePcm(samples);

// Transcribe from an in-memory WAV buffer
final bufText = await engine.transcribeWavBuffer(wavBytes);

// Clean up when done
engine.dispose();

Configuration

// Force a language (e.g. "English", "Chinese"); empty string = auto-detect
engine.setLanguage('English');

// Enable segmented mode for long audio (seconds per segment, 0 = off)
engine.setSegmentSec(30.0);

// Get timing stats from the last transcription
print(engine.perfStats());

Forced Alignment

Forced alignment (word-level timestamps for a known transcript) is available through the Rust library (qwen_asr::align::forced_align) for native integration. See the crate documentation for details.

Platform Support

Platform	Status
iOS	Supported
Android	Supported
macOS	Supported
Linux	Planned
Windows	Planned

License

MIT