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

Add Pascal API for Kokoro TTS models (#1724)

Browse Source
This commit is contained in:
Fangjun Kuang
2025-01-16 18:20:21 +08:00
committed by GitHub
parent 4335e2accd
commit 46f2e32e8a
10 changed files with 444 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
pascal-api-examples/tts/.gitignore vendored
View File

@@ -6,3 +6,5 @@ matcha-zh
matcha-en
matcha-zh-playback
matcha-en-playback
kokoro-en
kokoro-en-playback

239
pascal-api-examples/tts/kokoro-en-playback.pas Normal file
View File

@@ -0,0 +1,239 @@
{ Copyright (c) 2025 Xiaomi Corporation }
program kokoro_en_playback;
{
This file shows how to use the text to speech API of sherpa-onnx
with Kokoro models.
It generates speech from text and saves it to a wave file.
Note that it plays the audio back as it is still generating.
}
{$mode objfpc}
uses
{$ifdef unix}
cthreads,
{$endif}
SysUtils,
dos,
ctypes,
portaudio,
sherpa_onnx;
var
CriticalSection: TRTLCriticalSection;
Tts: TSherpaOnnxOfflineTts;
Audio: TSherpaOnnxGeneratedAudio;
Resampler: TSherpaOnnxLinearResampler;
Text: AnsiString;
Speed: Single = 1.0; {Use a larger value to speak faster}
SpeakerId: Integer = 7;
Buffer: TSherpaOnnxCircularBuffer;
FinishedGeneration: Boolean = False;
FinishedPlaying: Boolean = False;
Version: String;
EnvStr: String;
Status: Integer;
NumDevices: Integer;
DeviceIndex: Integer;
DeviceInfo: PPaDeviceInfo;
{ If you get EDivByZero: Division by zero error, please change the sample rate
to the one supported by your microphone.
}
DeviceSampleRate: Integer = 48000;
I: Integer;
Param: TPaStreamParameters;
Stream: PPaStream;
Wave: TSherpaOnnxWave;
function GenerateCallback(
Samples: pcfloat; N: cint32;
Arg: Pointer): cint; cdecl;
begin
EnterCriticalSection(CriticalSection);
try
if Resampler <> nil then
Buffer.Push(Resampler.Resample(Samples, N, False))
else
Buffer.Push(Samples, N);
finally
LeaveCriticalSection(CriticalSection);
end;
{ 1 means to continue generating; 0 means to stop generating. }
Result := 1;
end;
function PlayCallback(
input: Pointer; output: Pointer;
frameCount: culong;
timeInfo: PPaStreamCallbackTimeInfo;
statusFlags: TPaStreamCallbackFlags;
userData: Pointer ): cint; cdecl;
var
Samples: TSherpaOnnxSamplesArray;
I: Integer;
begin
EnterCriticalSection(CriticalSection);
try
if Buffer.Size >= frameCount then
begin
Samples := Buffer.Get(Buffer.Head, FrameCount);
Buffer.Pop(FrameCount);
end
else if Buffer.Size > 0 then
begin
Samples := Buffer.Get(Buffer.Head, Buffer.Size);
Buffer.Pop(Buffer.Size);
SetLength(Samples, frameCount);
end
else
SetLength(Samples, frameCount);
for I := 0 to frameCount - 1 do
pcfloat(output)[I] := Samples[I];
if (Buffer.Size > 0) or (not FinishedGeneration) then
Result := paContinue
else
begin
Result := paComplete;
FinishedPlaying := True;
end;
finally
LeaveCriticalSection(CriticalSection);
end;
end;
function GetOfflineTts: TSherpaOnnxOfflineTts;
var
Config: TSherpaOnnxOfflineTtsConfig;
begin
Config.Model.Kokoro.Model := './kokoro-en-v0_19/model.onnx';
Config.Model.Kokoro.Voices := './kokoro-en-v0_19/voices.bin';
Config.Model.Kokoro.Tokens := './kokoro-en-v0_19/tokens.txt';
Config.Model.Kokoro.DataDir := './kokoro-en-v0_19/espeak-ng-data';
Config.Model.NumThreads := 2;
Config.Model.Debug := False;
Config.MaxNumSentences := 1;
Result := TSherpaOnnxOfflineTts.Create(Config);
end;
begin
Tts := GetOfflineTts;
if Tts.GetSampleRate <> DeviceSampleRate then
Resampler := TSherpaOnnxLinearResampler.Create(Tts.GetSampleRate, DeviceSampleRate);
Version := String(Pa_GetVersionText);
WriteLn('Version is ', Version);
Status := Pa_Initialize;
if Status <> paNoError then
begin
WriteLn('Failed to initialize portaudio, ', Pa_GetErrorText(Status));
Exit;
end;
NumDevices := Pa_GetDeviceCount;
WriteLn('Num devices: ', NumDevices);
DeviceIndex := Pa_GetDefaultOutputDevice;
if DeviceIndex = paNoDevice then
begin
WriteLn('No default output device found');
Pa_Terminate;
Exit;
end;
EnvStr := GetEnv('SHERPA_ONNX_MIC_DEVICE');
if EnvStr <> '' then
begin
DeviceIndex := StrToIntDef(EnvStr, DeviceIndex);
WriteLn('Use device index from environment variable SHERPA_ONNX_MIC_DEVICE: ', EnvStr);
end;
for I := 0 to (NumDevices - 1) do
begin
DeviceInfo := Pa_GetDeviceInfo(I);
if I = DeviceIndex then
{ WriteLn(Format(' * %d %s', [I, DeviceInfo^.Name])) }
WriteLn(Format(' * %d %s', [I, AnsiString(DeviceInfo^.Name)]))
else
WriteLn(Format(' %d %s', [I, AnsiString(DeviceInfo^.Name)]));
end;
WriteLn('Use device ', DeviceIndex);
WriteLn(' Name ', Pa_GetDeviceInfo(DeviceIndex)^.Name);
WriteLn(' Max output channels ', Pa_GetDeviceInfo(DeviceIndex)^.MaxOutputChannels);
Initialize(Param);
Param.Device := DeviceIndex;
Param.ChannelCount := 1;
Param.SampleFormat := paFloat32;
param.SuggestedLatency := Pa_GetDeviceInfo(DeviceIndex)^.DefaultHighOutputLatency;
param.HostApiSpecificStreamInfo := nil;
Buffer := TSherpaOnnxCircularBuffer.Create(30 * DeviceSampleRate);
{ Note(fangjun): PortAudio invokes PlayCallback in a separate thread. }
Status := Pa_OpenStream(stream, nil, @Param, DeviceSampleRate, paFramesPerBufferUnspecified, paNoFlag,
PPaStreamCallback(@PlayCallback), nil);
if Status <> paNoError then
begin
WriteLn('Failed to open stream, ', Pa_GetErrorText(Status));
Pa_Terminate;
Exit;
end;
InitCriticalSection(CriticalSection);
Status := Pa_StartStream(stream);
if Status <> paNoError then
begin
WriteLn('Failed to start stream, ', Pa_GetErrorText(Status));
Pa_Terminate;
Exit;
end;
WriteLn('There are ', Tts.GetNumSpeakers, ' speakers');
Text := 'Friends fell out often because life was changing so fast. The easiest thing in the world was to lose touch with someone.';
Audio := Tts.Generate(Text, SpeakerId, Speed,
PSherpaOnnxGeneratedAudioCallbackWithArg(@GenerateCallback), nil);
FinishedGeneration := True;
SherpaOnnxWriteWave('./kokoro-en-playback-7.wav', Audio.Samples, Audio.SampleRate);
WriteLn('Saved to ./kokoro-en-playback-7.wav');
while not FinishedPlaying do
Pa_Sleep(100); {sleep for 0.1 second }
{TODO(fangjun): Use an event to indicate the play is finished}
DoneCriticalSection(CriticalSection);
FreeAndNil(Tts);
FreeAndNil(Resampler);
Status := Pa_CloseStream(stream);
if Status <> paNoError then
begin
WriteLn('Failed to close stream, ', Pa_GetErrorText(Status));
Exit;
end;
Status := Pa_Terminate;
if Status <> paNoError then
begin
WriteLn('Failed to deinitialize portaudio, ', Pa_GetErrorText(Status));
Exit;
end;
end.

55
pascal-api-examples/tts/kokoro-en.pas Normal file
View File

@@ -0,0 +1,55 @@
{ Copyright (c) 2025 Xiaomi Corporation }
program kokoro_en;
{
This file shows how to use the text to speech API of sherpa-onnx
with Kokoro TTS models.
It generates speech from text and saves it to a wave file.
If you want to play it while it is generating, please see
./kokoro-en-playback.pas
}
{$mode objfpc}
uses
SysUtils,
sherpa_onnx;
function GetOfflineTts: TSherpaOnnxOfflineTts;
var
Config: TSherpaOnnxOfflineTtsConfig;
begin
Config.Model.Kokoro.Model := './kokoro-en-v0_19/model.onnx';
Config.Model.Kokoro.Voices := './kokoro-en-v0_19/voices.bin';
Config.Model.Kokoro.Tokens := './kokoro-en-v0_19/tokens.txt';
Config.Model.Kokoro.DataDir := './kokoro-en-v0_19/espeak-ng-data';
Config.Model.NumThreads := 2;
Config.Model.Debug := False;
Config.MaxNumSentences := 1;
Result := TSherpaOnnxOfflineTts.Create(Config);
end;
var
Tts: TSherpaOnnxOfflineTts;
Audio: TSherpaOnnxGeneratedAudio;
Text: AnsiString;
Speed: Single = 1.0; {Use a larger value to speak faster}
SpeakerId: Integer = 8;
begin
Tts := GetOfflineTts;
WriteLn('There are ', Tts.GetNumSpeakers, ' speakers');
Text := 'Friends fell out often because life was changing so fast. The easiest thing in the world was to lose touch with someone.';
Audio := Tts.Generate(Text, SpeakerId, Speed);
SherpaOnnxWriteWave('./kokoro-en-8.wav', Audio.Samples, Audio.SampleRate);
WriteLn('Saved to ./kokoro-en-8.wav');
FreeAndNil(Tts);
end.

6
pascal-api-examples/tts/matcha-en-playback.pas
View File

@@ -2,7 +2,7 @@
program matcha_en_playback;
{
This file shows how to use the text to speech API of sherpa-onnx
with Piper models.
with MatchaTTS models.
It generates speech from text and saves it to a wave file.
@@ -210,8 +210,8 @@ begin
Audio := Tts.Generate(Text, SpeakerId, Speed,
PSherpaOnnxGeneratedAudioCallbackWithArg(@GenerateCallback), nil);
FinishedGeneration := True;
SherpaOnnxWriteWave('./matcha-zh-playback.wav', Audio.Samples, Audio.SampleRate);
WriteLn('Saved to ./matcha-zh-playback.wav');
SherpaOnnxWriteWave('./matcha-en-playback.wav', Audio.Samples, Audio.SampleRate);
WriteLn('Saved to ./matcha-en-playback.wav');
while not FinishedPlaying do
Pa_Sleep(100); {sleep for 0.1 second }

2
pascal-api-examples/tts/matcha-en.pas
View File

@@ -7,7 +7,7 @@ with MatchaTTS models.
It generates speech from text and saves it to a wave file.
If you want to play it while it is generating, please see
./matcha-zh-playback.pas
./matcha-en-playback.pas
}
{$mode objfpc}

2
pascal-api-examples/tts/matcha-zh-playback.pas
View File

@@ -2,7 +2,7 @@
program matcha_zh_playback;
{
This file shows how to use the text to speech API of sherpa-onnx
with Piper models.
with MatchaTTS models.
It generates speech from text and saves it to a wave file.

47
pascal-api-examples/tts/run-kokoro-en-playback.sh Executable file
View File

@@ -0,0 +1,47 @@
#!/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: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
popd
fi
# please visit
# https://k2-fsa.github.io/sherpa/onnx/tts/pretrained_models/kokoro.html
if [ ! -f ./kokoro-en-v0_19/model.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-en-v0_19.tar.bz2
tar xf kokoro-en-v0_19.tar.bz2
rm kokoro-en-v0_19.tar.bz2
fi
fpc \
-dSHERPA_ONNX_USE_SHARED_LIBS \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
-Fl/usr/local/Cellar/portaudio/19.7.0/lib \
./kokoro-en-playback.pas
# Please see ../portaudio-test/README.md
# for how to install portaudio on macOS
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./kokoro-en-playback

43
pascal-api-examples/tts/run-kokoro-en.sh Executable file
View File

@@ -0,0 +1,43 @@
#!/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: $SHERPA_ONNX_DIR"
if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
mkdir -p ../../build
pushd ../../build
cmake \
-DCMAKE_INSTALL_PREFIX=./install \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
..
cmake --build . --target install --config Release
popd
fi
# please visit
# https://k2-fsa.github.io/sherpa/onnx/tts/pretrained_models/kokoro.html
if [ ! -f ./kokoro-en-v0_19/model.onnx ]; then
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-en-v0_19.tar.bz2
tar xf kokoro-en-v0_19.tar.bz2
rm kokoro-en-v0_19.tar.bz2
fi
fpc \
-dSHERPA_ONNX_USE_SHARED_LIBS \
-Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
-Fl$SHERPA_ONNX_DIR/build/install/lib \
./kokoro-en.pas
export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
./kokoro-en