EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Pascal API for VAD (#1249)

Browse Source
This commit is contained in:
Fangjun Kuang
2024-08-13 16:16:51 +08:00
committed by GitHub
parent a7dc6c2c16
commit 619279b162
24 changed files with 1199 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
pascal-api-examples/vad-with-non-streaming-asr/.gitignore vendored Normal file
View File

@@ -0,0 +1,3 @@
!run-*.sh
vad_with_whisper
vad_with_sense_voice

12
pascal-api-examples/vad-with-non-streaming-asr/README.md Normal file
View File

@@ -0,0 +1,12 @@
# Introduction
This directory contains examples for how to use the VAD (voice activity detection)
with non-streaming speech recognition models.
|Directory| Description|
|---------|------------|
|[run-vad-with-whisper.sh](./run-vad-with-whisper.sh)|It shows how to use the VAD + Whisper for speech recognition.|
|[run-vad-with-sense-voice.sh](./run-vad-with-sense-voice.sh)|It shows how to use the VAD + SenseVoice for speech recognition.|
Please refer to [non-streaming-asr](../non-streaming-asr) for more kinds of non-streaming models.

48
pascal-api-examples/vad-with-non-streaming-asr/run-vad-with-sense-voice.sh Executable file
View File

@@ -0,0 +1,48 @@
#!/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: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
popd
fi
if [[ ! -f ./silero_vad.onnx ]]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
fi
if [ ! -f ./lei-jun-test.wav ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/lei-jun-test.wav
fi
if [ ! -f ./sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/tokens.txt ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
tar xvf sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
rm sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17.tar.bz2
fi
fpc \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
./vad_with_sense_voice.pas
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./vad_with_sense_voice

49
pascal-api-examples/vad-with-non-streaming-asr/run-vad-with-whisper.sh Executable file
View File

@@ -0,0 +1,49 @@
#!/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: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
popd
fi
if [[ ! -f ./silero_vad.onnx ]]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
fi
if [ ! -f ./Obama.wav ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/Obama.wav
fi
if [ ! -f ./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-whisper-tiny.en.tar.bz2
tar xvf sherpa-onnx-whisper-tiny.en.tar.bz2
rm sherpa-onnx-whisper-tiny.en.tar.bz2
fi
fpc \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
./vad_with_whisper.pas
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./vad_with_whisper

137
pascal-api-examples/vad-with-non-streaming-asr/vad_with_sense_voice.pas Normal file
View File

@@ -0,0 +1,137 @@
{ Copyright (c) 2024 Xiaomi Corporation }
{
This file shows how to use a non-streaming SenseVoice model
with silero VAD to decode files.
You can download the model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
}
program vad_with_whisper;
{$mode objfpc}
uses
sherpa_onnx,
SysUtils;
function CreateVad(): TSherpaOnnxVoiceActivityDetector;
var
Config: TSherpaOnnxVadModelConfig;
SampleRate: Integer;
WindowSize: Integer;
begin
Initialize(Config);
SampleRate := 16000; {Please don't change it unless you know the details}
WindowSize := 512; {Please don't change it unless you know the details}
Config.SileroVad.Model := './silero_vad.onnx';
Config.SileroVad.MinSpeechDuration := 0.5;
Config.SileroVad.MinSilenceDuration := 0.5;
Config.SileroVad.Threshold := 0.5;
Config.SileroVad.WindowSize := WindowSize;
Config.NumThreads:= 1;
Config.Debug:= True;
Config.Provider:= 'cpu';
Config.SampleRate := SampleRate;
Result := TSherpaOnnxVoiceActivityDetector.Create(Config, 30);
end;
function CreateOfflineRecognizer(): TSherpaOnnxOfflineRecognizer;
var
Config: TSherpaOnnxOfflineRecognizerConfig;
begin
Initialize(Config);
Config.ModelConfig.SenseVoice.Model := './sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/model.int8.onnx';
Config.ModelConfig.SenseVoice.Language := 'auto';
Config.ModelConfig.SenseVoice.UseItn := False;
Config.ModelConfig.Tokens := './sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17/tokens.txt';
Config.ModelConfig.Provider := 'cpu';
Config.ModelConfig.NumThreads := 1;
Config.ModelConfig.Debug := False;
Result := TSherpaOnnxOfflineRecognizer.Create(Config);
end;
var
Wave: TSherpaOnnxWave;
Recognizer: TSherpaOnnxOfflineRecognizer;
Vad: TSherpaOnnxVoiceActivityDetector;
Offset: Integer;
WindowSize: Integer;
SpeechSegment: TSherpaOnnxSpeechSegment;
Start: Single;
Duration: Single;
Stream: TSherpaOnnxOfflineStream;
RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
begin
Vad := CreateVad();
Recognizer := CreateOfflineRecognizer();
Wave := SherpaOnnxReadWave('./lei-jun-test.wav');
if Wave.SampleRate <> Vad.Config.SampleRate then
begin
WriteLn(Format('Expected sample rate: %d. Given: %d',
[Vad.Config.SampleRate, Wave.SampleRate]));
Exit;
end;
WindowSize := Vad.Config.SileroVad.WindowSize;
Offset := 0;
while Offset + WindowSize <= Length(Wave.Samples) do
begin
Vad.AcceptWaveform(Wave.Samples, Offset, WindowSize);
Offset += WindowSize;
while not Vad.IsEmpty do
begin
SpeechSegment := Vad.Front();
Vad.Pop();
Stream := Recognizer.CreateStream();
Stream.AcceptWaveform(SpeechSegment.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Start := SpeechSegment.Start / Wave.SampleRate;
Duration := Length(SpeechSegment.Samples) / Wave.SampleRate;
WriteLn(Format('%.3f -- %.3f %s',
[Start, Start + Duration, RecognitionResult.Text]));
FreeAndNil(Stream);
end;
end;
Vad.Flush;
while not Vad.IsEmpty do
begin
SpeechSegment := Vad.Front();
Vad.Pop();
Stream := Recognizer.CreateStream();
Stream.AcceptWaveform(SpeechSegment.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Start := SpeechSegment.Start / Wave.SampleRate;
Duration := Length(SpeechSegment.Samples) / Wave.SampleRate;
WriteLn(Format('%.3f -- %.3f %s',
[Start, Start + Duration, RecognitionResult.Text]));
FreeAndNil(Stream);
end;
FreeAndNil(Recognizer);
FreeAndNil(Vad);
end.

136
pascal-api-examples/vad-with-non-streaming-asr/vad_with_whisper.pas Normal file
View File

@@ -0,0 +1,136 @@
{ Copyright (c) 2024 Xiaomi Corporation }
{
This file shows how to use a non-streaming Whisper model
with silero VAD to decode files.
You can download the model files from
https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
}
program vad_with_whisper;
{$mode objfpc}
uses
sherpa_onnx,
SysUtils;
function CreateVad(): TSherpaOnnxVoiceActivityDetector;
var
Config: TSherpaOnnxVadModelConfig;
SampleRate: Integer;
WindowSize: Integer;
begin
Initialize(Config);
SampleRate := 16000; {Please don't change it unless you know the details}
WindowSize := 512; {Please don't change it unless you know the details}
Config.SileroVad.Model := './silero_vad.onnx';
Config.SileroVad.MinSpeechDuration := 0.5;
Config.SileroVad.MinSilenceDuration := 0.5;
Config.SileroVad.Threshold := 0.5;
Config.SileroVad.WindowSize := WindowSize;
Config.NumThreads:= 1;
Config.Debug:= True;
Config.Provider:= 'cpu';
Config.SampleRate := SampleRate;
Result := TSherpaOnnxVoiceActivityDetector.Create(Config, 30);
end;
function CreateOfflineRecognizer(): TSherpaOnnxOfflineRecognizer;
var
Config: TSherpaOnnxOfflineRecognizerConfig;
begin
Initialize(Config);
Config.ModelConfig.Whisper.Encoder := './sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx';
Config.ModelConfig.Whisper.Decoder := './sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx';
Config.ModelConfig.Tokens := './sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt';
Config.ModelConfig.Provider := 'cpu';
Config.ModelConfig.NumThreads := 1;
Config.ModelConfig.Debug := False;
Result := TSherpaOnnxOfflineRecognizer.Create(Config);
end;
var
Wave: TSherpaOnnxWave;
Recognizer: TSherpaOnnxOfflineRecognizer;
Vad: TSherpaOnnxVoiceActivityDetector;
Offset: Integer;
WindowSize: Integer;
SpeechSegment: TSherpaOnnxSpeechSegment;
Start: Single;
Duration: Single;
Stream: TSherpaOnnxOfflineStream;
RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
begin
Vad := CreateVad();
Recognizer := CreateOfflineRecognizer();
Wave := SherpaOnnxReadWave('./Obama.wav');
if Wave.SampleRate <> Vad.Config.SampleRate then
begin
WriteLn(Format('Expected sample rate: %d. Given: %d',
[Vad.Config.SampleRate, Wave.SampleRate]));
Exit;
end;
WindowSize := Vad.Config.SileroVad.WindowSize;
Offset := 0;
while Offset + WindowSize <= Length(Wave.Samples) do
begin
Vad.AcceptWaveform(Wave.Samples, Offset, WindowSize);
Offset += WindowSize;
while not Vad.IsEmpty do
begin
SpeechSegment := Vad.Front();
Vad.Pop();
Stream := Recognizer.CreateStream();
Stream.AcceptWaveform(SpeechSegment.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Start := SpeechSegment.Start / Wave.SampleRate;
Duration := Length(SpeechSegment.Samples) / Wave.SampleRate;
WriteLn(Format('%.3f -- %.3f %s',
[Start, Start + Duration, RecognitionResult.Text]));
FreeAndNil(Stream);
end;
end;
Vad.Flush;
while not Vad.IsEmpty do
begin
SpeechSegment := Vad.Front();
Vad.Pop();
Stream := Recognizer.CreateStream();
Stream.AcceptWaveform(SpeechSegment.Samples, Wave.SampleRate);
Recognizer.Decode(Stream);
RecognitionResult := Recognizer.GetResult(Stream);
Start := SpeechSegment.Start / Wave.SampleRate;
Duration := Length(SpeechSegment.Samples) / Wave.SampleRate;
WriteLn(Format('%.3f -- %.3f %s',
[Start, Start + Duration, RecognitionResult.Text]));
FreeAndNil(Stream);
end;
FreeAndNil(Recognizer);
FreeAndNil(Vad);
end.