Add Golang API for spoken language identification. (#709)
This commit is contained in:
@@ -23,4 +23,7 @@ for details.
|
|||||||
- [./vad-asr-paraformer](./vad-asr-paraformer) It shows how to use silero VAD + Paraformer
|
- [./vad-asr-paraformer](./vad-asr-paraformer) It shows how to use silero VAD + Paraformer
|
||||||
for speech recognition.
|
for speech recognition.
|
||||||
|
|
||||||
|
- [./vad-spoken-language-identification](./vad-spoken-language-identification) It shows how to use silero VAD + Whisper
|
||||||
|
for spoken language identification.
|
||||||
|
|
||||||
[sherpa-onnx]: https://github.com/k2-fsa/sherpa-onnx
|
[sherpa-onnx]: https://github.com/k2-fsa/sherpa-onnx
|
||||||
|
|||||||
@@ -0,0 +1,3 @@
|
|||||||
|
module vad-spoken-language-identification
|
||||||
|
|
||||||
|
go 1.12
|
||||||
141
go-api-examples/vad-spoken-language-identification/main.go
Normal file
141
go-api-examples/vad-spoken-language-identification/main.go
Normal file
@@ -0,0 +1,141 @@
|
|||||||
|
package main
|
||||||
|
|
||||||
|
import (
|
||||||
|
"fmt"
|
||||||
|
iso639 "github.com/barbashov/iso639-3"
|
||||||
|
"github.com/gordonklaus/portaudio"
|
||||||
|
sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
|
||||||
|
"log"
|
||||||
|
)
|
||||||
|
|
||||||
|
func main() {
|
||||||
|
log.SetFlags(log.LstdFlags | log.Lmicroseconds)
|
||||||
|
|
||||||
|
// 1. Create VAD
|
||||||
|
config := sherpa.VadModelConfig{}
|
||||||
|
|
||||||
|
// Please download silero_vad.onnx from
|
||||||
|
// https://github.com/snakers4/silero-vad/blob/master/files/silero_vad.onnx
|
||||||
|
|
||||||
|
config.SileroVad.Model = "./silero_vad.onnx"
|
||||||
|
config.SileroVad.Threshold = 0.5
|
||||||
|
config.SileroVad.MinSilenceDuration = 0.5
|
||||||
|
config.SileroVad.MinSpeechDuration = 0.25
|
||||||
|
config.SileroVad.WindowSize = 512
|
||||||
|
config.SampleRate = 16000
|
||||||
|
config.NumThreads = 1
|
||||||
|
config.Provider = "cpu"
|
||||||
|
config.Debug = 1
|
||||||
|
|
||||||
|
var bufferSizeInSeconds float32 = 20
|
||||||
|
|
||||||
|
vad := sherpa.NewVoiceActivityDetector(&config, bufferSizeInSeconds)
|
||||||
|
defer sherpa.DeleteVoiceActivityDetector(vad)
|
||||||
|
|
||||||
|
// 2. Create spoken language identifier
|
||||||
|
|
||||||
|
c := sherpa.SpokenLanguageIdentificationConfig{}
|
||||||
|
c.Whisper.Encoder = "./sherpa-onnx-whisper-tiny/tiny-encoder.int8.onnx"
|
||||||
|
c.Whisper.Decoder = "./sherpa-onnx-whisper-tiny/tiny-decoder.int8.onnx"
|
||||||
|
c.NumThreads = 2
|
||||||
|
c.Debug = 1
|
||||||
|
c.Provider = "cpu"
|
||||||
|
|
||||||
|
slid := sherpa.NewSpokenLanguageIdentification(&c)
|
||||||
|
defer sherpa.DeleteSpokenLanguageIdentification(slid)
|
||||||
|
|
||||||
|
err := portaudio.Initialize()
|
||||||
|
if err != nil {
|
||||||
|
log.Fatalf("Unable to initialize portaudio: %v\n", err)
|
||||||
|
}
|
||||||
|
defer portaudio.Terminate()
|
||||||
|
|
||||||
|
default_device, err := portaudio.DefaultInputDevice()
|
||||||
|
if err != nil {
|
||||||
|
log.Fatal("Failed to get default input device: %v\n", err)
|
||||||
|
}
|
||||||
|
log.Printf("Selected default input device: %s\n", default_device.Name)
|
||||||
|
param := portaudio.StreamParameters{}
|
||||||
|
param.Input.Device = default_device
|
||||||
|
param.Input.Channels = 1
|
||||||
|
param.Input.Latency = default_device.DefaultHighInputLatency
|
||||||
|
|
||||||
|
param.SampleRate = float64(config.SampleRate)
|
||||||
|
param.FramesPerBuffer = 0
|
||||||
|
param.Flags = portaudio.ClipOff
|
||||||
|
|
||||||
|
// you can choose another value for 0.1 if you want
|
||||||
|
samplesPerCall := int32(param.SampleRate * 0.1) // 0.1 second
|
||||||
|
samples := make([]float32, samplesPerCall)
|
||||||
|
|
||||||
|
s, err := portaudio.OpenStream(param, samples)
|
||||||
|
if err != nil {
|
||||||
|
log.Fatalf("Failed to open the stream")
|
||||||
|
}
|
||||||
|
|
||||||
|
defer s.Close()
|
||||||
|
chk(s.Start())
|
||||||
|
|
||||||
|
log.Print("Started! Please speak")
|
||||||
|
printed := false
|
||||||
|
|
||||||
|
k := 0
|
||||||
|
for {
|
||||||
|
chk(s.Read())
|
||||||
|
vad.AcceptWaveform(samples)
|
||||||
|
|
||||||
|
if vad.IsSpeech() && !printed {
|
||||||
|
printed = true
|
||||||
|
log.Print("Detected speech\n")
|
||||||
|
}
|
||||||
|
|
||||||
|
if !vad.IsSpeech() {
|
||||||
|
printed = false
|
||||||
|
}
|
||||||
|
|
||||||
|
for !vad.IsEmpty() {
|
||||||
|
speechSegment := vad.Front()
|
||||||
|
vad.Pop()
|
||||||
|
|
||||||
|
duration := float32(len(speechSegment.Samples)) / float32(config.SampleRate)
|
||||||
|
|
||||||
|
audio := &sherpa.GeneratedAudio{}
|
||||||
|
audio.Samples = speechSegment.Samples
|
||||||
|
audio.SampleRate = config.SampleRate
|
||||||
|
|
||||||
|
// Now decode it
|
||||||
|
go decode(slid, audio, k)
|
||||||
|
|
||||||
|
k += 1
|
||||||
|
|
||||||
|
log.Printf("Duration: %.2f seconds\n", duration)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
chk(s.Stop())
|
||||||
|
}
|
||||||
|
|
||||||
|
func decode(slid *sherpa.SpokenLanguageIdentification, audio *sherpa.GeneratedAudio, id int) {
|
||||||
|
stream := slid.CreateStream()
|
||||||
|
defer sherpa.DeleteOfflineStream(stream)
|
||||||
|
|
||||||
|
stream.AcceptWaveform(audio.SampleRate, audio.Samples)
|
||||||
|
result := slid.Compute(stream)
|
||||||
|
lang := iso639.FromPart1Code(result.Lang).Name
|
||||||
|
log.Printf("Detected language: %v", lang)
|
||||||
|
|
||||||
|
duration := float32(len(audio.Samples)) / float32(audio.SampleRate)
|
||||||
|
|
||||||
|
filename := fmt.Sprintf("seg-%d-%.2f-seconds-%s.wav", id, duration, lang)
|
||||||
|
ok := audio.Save(filename)
|
||||||
|
if ok {
|
||||||
|
log.Printf("Saved to %s", filename)
|
||||||
|
}
|
||||||
|
log.Print("----------\n")
|
||||||
|
}
|
||||||
|
|
||||||
|
func chk(err error) {
|
||||||
|
if err != nil {
|
||||||
|
panic(err)
|
||||||
|
}
|
||||||
|
}
|
||||||
16
go-api-examples/vad-spoken-language-identification/run.sh
Executable file
16
go-api-examples/vad-spoken-language-identification/run.sh
Executable file
@@ -0,0 +1,16 @@
|
|||||||
|
#!/usr/bin/env bash
|
||||||
|
|
||||||
|
|
||||||
|
if [ ! -f ./silero_vad.onnx ]; then
|
||||||
|
curl -SL -O https://github.com/snakers4/silero-vad/blob/master/files/silero_vad.onnx
|
||||||
|
fi
|
||||||
|
|
||||||
|
if [ ! -f ./sherpa-onnx-whisper-tiny/tiny-encoder.int8.onnx ]; then
|
||||||
|
curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-whisper-tiny.tar.bz2
|
||||||
|
tar xvf sherpa-onnx-whisper-tiny.tar.bz2
|
||||||
|
rm sherpa-onnx-whisper-tiny.tar.bz2
|
||||||
|
fi
|
||||||
|
|
||||||
|
go mod tidy
|
||||||
|
go build
|
||||||
|
./vad-spoken-language-identification
|
||||||
2
scripts/go/_internal/vad-spoken-language-identification/.gitignore
vendored
Normal file
2
scripts/go/_internal/vad-spoken-language-identification/.gitignore
vendored
Normal file
@@ -0,0 +1,2 @@
|
|||||||
|
vad-spoken-language-identification
|
||||||
|
|
||||||
@@ -0,0 +1,5 @@
|
|||||||
|
module vad-spoken-language-identification
|
||||||
|
|
||||||
|
go 1.12
|
||||||
|
|
||||||
|
replace github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx => ../
|
||||||
1
scripts/go/_internal/vad-spoken-language-identification/main.go
Symbolic link
1
scripts/go/_internal/vad-spoken-language-identification/main.go
Symbolic link
@@ -0,0 +1 @@
|
|||||||
|
/Users/fangjun/open-source/sherpa-onnx/go-api-examples/vad-spoken-language-identification/main.go
|
||||||
1
scripts/go/_internal/vad-spoken-language-identification/run.sh
Symbolic link
1
scripts/go/_internal/vad-spoken-language-identification/run.sh
Symbolic link
@@ -0,0 +1 @@
|
|||||||
|
/Users/fangjun/open-source/sherpa-onnx/go-api-examples/vad-spoken-language-identification/run.sh
|
||||||
@@ -783,3 +783,72 @@ func (vad *VoiceActivityDetector) Front() *SpeechSegment {
|
|||||||
func (vad *VoiceActivityDetector) Reset() {
|
func (vad *VoiceActivityDetector) Reset() {
|
||||||
C.SherpaOnnxVoiceActivityDetectorReset(vad.impl)
|
C.SherpaOnnxVoiceActivityDetectorReset(vad.impl)
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// Spoken language identification
|
||||||
|
|
||||||
|
type SpokenLanguageIdentificationWhisperConfig struct {
|
||||||
|
Encoder string
|
||||||
|
Decoder string
|
||||||
|
TailPaddings int
|
||||||
|
}
|
||||||
|
|
||||||
|
type SpokenLanguageIdentificationConfig struct {
|
||||||
|
Whisper SpokenLanguageIdentificationWhisperConfig
|
||||||
|
NumThreads int
|
||||||
|
Debug int
|
||||||
|
Provider string
|
||||||
|
}
|
||||||
|
|
||||||
|
type SpokenLanguageIdentification struct {
|
||||||
|
impl *C.struct_SherpaOnnxSpokenLanguageIdentification
|
||||||
|
}
|
||||||
|
|
||||||
|
type SpokenLanguageIdentificationResult struct {
|
||||||
|
Lang string
|
||||||
|
}
|
||||||
|
|
||||||
|
func NewSpokenLanguageIdentification(config *SpokenLanguageIdentificationConfig) *SpokenLanguageIdentification {
|
||||||
|
c := C.struct_SherpaOnnxSpokenLanguageIdentificationConfig{}
|
||||||
|
|
||||||
|
c.whisper.encoder = C.CString(config.Whisper.Encoder)
|
||||||
|
defer C.free(unsafe.Pointer(c.whisper.encoder))
|
||||||
|
|
||||||
|
c.whisper.decoder = C.CString(config.Whisper.Decoder)
|
||||||
|
defer C.free(unsafe.Pointer(c.whisper.decoder))
|
||||||
|
|
||||||
|
c.whisper.tail_paddings = C.int(config.Whisper.TailPaddings)
|
||||||
|
|
||||||
|
c.num_threads = C.int(config.NumThreads)
|
||||||
|
c.debug = C.int(config.Debug)
|
||||||
|
|
||||||
|
c.provider = C.CString(config.Provider)
|
||||||
|
defer C.free(unsafe.Pointer(c.provider))
|
||||||
|
|
||||||
|
slid := &SpokenLanguageIdentification{}
|
||||||
|
slid.impl = C.SherpaOnnxCreateSpokenLanguageIdentification(&c)
|
||||||
|
|
||||||
|
return slid
|
||||||
|
}
|
||||||
|
|
||||||
|
func DeleteSpokenLanguageIdentification(slid *SpokenLanguageIdentification) {
|
||||||
|
C.SherpaOnnxDestroySpokenLanguageIdentification(slid.impl)
|
||||||
|
slid.impl = nil
|
||||||
|
}
|
||||||
|
|
||||||
|
// The user has to invoke DeleteOfflineStream() to free the returned value
|
||||||
|
// to avoid memory leak
|
||||||
|
func (slid *SpokenLanguageIdentification) CreateStream() *OfflineStream {
|
||||||
|
stream := &OfflineStream{}
|
||||||
|
stream.impl = C.SherpaOnnxSpokenLanguageIdentificationCreateOfflineStream(slid.impl)
|
||||||
|
return stream
|
||||||
|
}
|
||||||
|
|
||||||
|
func (slid *SpokenLanguageIdentification) Compute(stream *OfflineStream) *SpokenLanguageIdentificationResult {
|
||||||
|
r := C.SherpaOnnxSpokenLanguageIdentificationCompute(slid.impl, stream.impl)
|
||||||
|
// defer C.SherpaOnnxDestroySpokenLanguageIdentificationResult(r)
|
||||||
|
|
||||||
|
ans := &SpokenLanguageIdentificationResult{}
|
||||||
|
ans.Lang = C.GoString(r.lang)
|
||||||
|
|
||||||
|
return ans
|
||||||
|
}
|
||||||
|
|||||||
@@ -91,7 +91,7 @@ std::string SpokenLanguageIdentificationConfig::ToString() const {
|
|||||||
std::ostringstream os;
|
std::ostringstream os;
|
||||||
|
|
||||||
os << "SpokenLanguageIdentificationConfig(";
|
os << "SpokenLanguageIdentificationConfig(";
|
||||||
os << "whisper=\"" << whisper.ToString() << "\", ";
|
os << "whisper=" << whisper.ToString() << ", ";
|
||||||
os << "num_threads=" << num_threads << ", ";
|
os << "num_threads=" << num_threads << ", ";
|
||||||
os << "debug=" << (debug ? "True" : "False") << ", ";
|
os << "debug=" << (debug ? "True" : "False") << ", ";
|
||||||
os << "provider=\"" << provider << "\")";
|
os << "provider=\"" << provider << "\")";
|
||||||
|
|||||||
Reference in New Issue
Block a user