Add Pascal/Go/C#/Dart API for NeMo Canary ASR models (#2367)

Add support for the new NeMo Canary ASR model across multiple language bindings by introducing a Canary model configuration and setter method on the offline recognizer.

- Define Canary model config in Pascal, Go, C#, Dart and update converter functions
- Add SetConfig API for offline recognizer (Pascal, Go, C#, Dart)
- Extend CI/workflows and example scripts to test non-streaming Canary decoding
This commit is contained in:
Fangjun Kuang
2025-07-10 14:53:33 +08:00
committed by GitHub
parent e2b2d5ea57
commit fd9a687ec2
27 changed files with 779 additions and 8 deletions

View File

@@ -0,0 +1,84 @@
// Copyright (c) 2025 Xiaomi Corporation
import 'dart:io';
import 'package:args/args.dart';
import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
import './init.dart';
void main(List<String> arguments) async {
await initSherpaOnnx();
final parser = ArgParser()
..addOption('encoder', help: 'Path to the NeMo Canary encoder model')
..addOption('decoder', help: 'Path to the NeMo Canary decoder model')
..addOption('src-lang', help: 'Language of the input audio')
..addOption('tgt-lang', help: 'Language of the recognition result')
..addOption('tokens', help: 'Path to tokens.txt')
..addOption('input-wav', help: 'Path to input.wav to transcribe');
final res = parser.parse(arguments);
if (res['encoder'] == null ||
res['decoder'] == null ||
res['src-lang'] == null ||
res['tgt-lang'] == null ||
res['tokens'] == null ||
res['input-wav'] == null) {
print(parser.usage);
exit(1);
}
final encoder = res['encoder'] as String;
final decoder = res['decoder'] as String;
final srcLang = res['src-lang'] as String;
final tgtLang = res['tgt-lang'] as String;
final tokens = res['tokens'] as String;
final inputWav = res['input-wav'] as String;
final canary = sherpa_onnx.OfflineCanaryModelConfig(
encoder: encoder, decoder: decoder, srcLang: srcLang, tgtLang: tgtLang);
final modelConfig = sherpa_onnx.OfflineModelConfig(
canary: canary,
tokens: tokens,
debug: false,
numThreads: 1,
);
var config = sherpa_onnx.OfflineRecognizerConfig(model: modelConfig);
final recognizer = sherpa_onnx.OfflineRecognizer(config);
final waveData = sherpa_onnx.readWave(inputWav);
final stream = recognizer.createStream();
stream.acceptWaveform(
samples: waveData.samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
print('Result in $tgtLang: ${result.text}');
stream.free();
// Example to change the target language to de
if (tgtLang != 'en') {
var json = config.toJson();
((json['model'] as Map<String, dynamic>)!['canary']
as Map<String, dynamic>)!['tgtLang'] = 'en';
config = sherpa_onnx.OfflineRecognizerConfig.fromJson(json);
recognizer.setConfig(config);
final stream = recognizer.createStream();
stream.acceptWaveform(
samples: waveData.samples, sampleRate: waveData.sampleRate);
recognizer.decode(stream);
final result = recognizer.getResult(stream);
print('Result in English: ${result.text}');
stream.free();
}
recognizer.free();
}

View File

@@ -0,0 +1,33 @@
#!/usr/bin/env bash
set -ex
dart pub get
if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
fi
for tgt_lang in en de es fr; do
dart run \
./bin/nemo-canary.dart \
--encoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx \
--decoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx \
--tokens ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt \
--src-lang en \
--tgt-lang $tgt_lang \
--input-wav ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav
done
for tgt_lang in en de; do
dart run \
./bin/nemo-canary.dart \
--encoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx \
--decoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx \
--tokens ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt \
--src-lang de \
--tgt-lang $tgt_lang \
--input-wav ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/de.wav
done