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

@@ -10,3 +10,4 @@ telespeech_ctc
moonshine
dolphin_ctc
zipformer_ctc
nemo_canary

View File

@@ -0,0 +1,107 @@
{ Copyright (c) 2025 Xiaomi Corporation }
{
This file shows how to use a non-streaming NeMo Canary model
to decode files.
You can download the model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
}
program nemo_canary;
{$mode objfpc}
uses
sherpa_onnx,
DateUtils,
SysUtils;
var
Wave: TSherpaOnnxWave;
WaveFilename: AnsiString;
Config: TSherpaOnnxOfflineRecognizerConfig;
Recognizer: TSherpaOnnxOfflineRecognizer;
Stream: TSherpaOnnxOfflineStream;
RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
Start: TDateTime;
Stop: TDateTime;
Elapsed: Single;
Duration: Single;
RealTimeFactor: Single;
begin
Initialize(Config);
Config.ModelConfig.Canary.Encoder := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx';
Config.ModelConfig.Canary.Decoder := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx';
Config.ModelConfig.Canary.SrcLang := 'en';
Config.ModelConfig.Canary.TgtLang := 'en';
Config.ModelConfig.Canary.UsePnc := True;
Config.ModelConfig.Tokens := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt';
Config.ModelConfig.Provider := 'cpu';
Config.ModelConfig.NumThreads := 1;
Config.ModelConfig.Debug := False;
WaveFilename := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav';
Wave := SherpaOnnxReadWave(WaveFilename);
Recognizer := TSherpaOnnxOfflineRecognizer.Create(Config);
Stream := Recognizer.CreateStream();
Start := Now;
Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Stop := Now;
Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
Duration := Length(Wave.Samples) / Wave.SampleRate;
RealTimeFactor := Elapsed / Duration;
WriteLn(RecognitionResult.ToString);
WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
WriteLn(Format('Elapsed %.3f s', [Elapsed]));
WriteLn(Format('Wave duration %.3f s', [Duration]));
WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
FreeAndNil(Stream);
WriteLn('-----------Output German-----');
Stream := Recognizer.CreateStream();
Start := Now;
Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
Config.ModelConfig.Canary.TgtLang := 'de';
Recognizer.SetConfig(Config);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Stop := Now;
Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
Duration := Length(Wave.Samples) / Wave.SampleRate;
RealTimeFactor := Elapsed / Duration;
WriteLn(RecognitionResult.ToString);
WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
WriteLn(Format('Elapsed %.3f s', [Elapsed]));
WriteLn(Format('Wave duration %.3f s', [Duration]));
WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
{Free resources to avoid memory leak.
Note: You don't need to invoke them for this simple script.
However, you have to invoke them in your own large/complex project.
}
FreeAndNil(Stream);
FreeAndNil(Recognizer);
end.

View File

@@ -0,0 +1,42 @@
#!/usr/bin/env bash
set -ex
SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
SHERPA_ONNX_DIR=$(cd $SCRIPT_DIR/../.. && pwd)
echo "SHERPA_ONNX_DIR: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
ls -lh lib
popd
fi
if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
fi
fpc \
-dSHERPA_ONNX_USE_SHARED_LIBS \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
./nemo_canary.pas
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./nemo_canary