Fangjun Kuang
Committed by GitHub

Support decoding multiple streams in Java API. (#2149)

... ... @@ -166,6 +166,8 @@ jobs:
rm -rf sherpa-onnx-fire-red-*
./run-non-streaming-decode-file-whisper.sh
./run-non-streaming-decode-file-whisper-multiple.sh
rm -rf sherpa-onnx-whisper-*
./run-non-streaming-decode-file-nemo.sh
... ...
// Copyright 2025 Xiaomi Corporation
// This file shows how to use an offline whisper, i.e., non-streaming whisper,
// to decode files.
import com.k2fsa.sherpa.onnx.*;
public class NonStreamingDecodeFileWhisperMultiple {
public static void main(String[] args) {
// please refer to
// https://k2-fsa.github.io/sherpa/onnx/pretrained_models/whisper/tiny.en.html
// to download model files
String encoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx";
String decoder = "./sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx";
String tokens = "./sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt";
String waveFilename0 = "./sherpa-onnx-whisper-tiny.en/test_wavs/0.wav";
String waveFilename1 = "./sherpa-onnx-whisper-tiny.en/test_wavs/1.wav";
WaveReader reader0 = new WaveReader(waveFilename0);
WaveReader reader1 = new WaveReader(waveFilename1);
OfflineWhisperModelConfig whisper =
OfflineWhisperModelConfig.builder().setEncoder(encoder).setDecoder(decoder).build();
OfflineModelConfig modelConfig =
OfflineModelConfig.builder()
.setWhisper(whisper)
.setTokens(tokens)
.setNumThreads(1)
.setDebug(true)
.build();
OfflineRecognizerConfig config =
OfflineRecognizerConfig.builder()
.setOfflineModelConfig(modelConfig)
.setDecodingMethod("greedy_search")
.build();
OfflineRecognizer recognizer = new OfflineRecognizer(config);
OfflineStream stream0 = recognizer.createStream();
stream0.acceptWaveform(reader0.getSamples(), reader0.getSampleRate());
OfflineStream stream1 = recognizer.createStream();
stream1.acceptWaveform(reader1.getSamples(), reader1.getSampleRate());
OfflineStream[] ss = new OfflineStream[] {stream0, stream1};
recognizer.decode(ss);
String text0 = recognizer.getResult(stream0).getText();
String text1 = recognizer.getResult(stream1).getText();
System.out.printf("filename0:%s\nresult0:%s\n\n", waveFilename0, text0);
System.out.printf("filename1:%s\nresult1:%s\n\n", waveFilename1, text1);
stream0.release();
stream1.release();
recognizer.release();
}
}
... ...
#!/usr/bin/env bash
set -ex
if [[ ! -f ../build/lib/libsherpa-onnx-jni.dylib && ! -f ../build/lib/libsherpa-onnx-jni.so ]]; then
mkdir -p ../build
pushd ../build
cmake \
-DSHERPA_ONNX_ENABLE_PYTHON=OFF \
-DSHERPA_ONNX_ENABLE_TESTS=OFF \
-DSHERPA_ONNX_ENABLE_CHECK=OFF \
-DBUILD_SHARED_LIBS=ON \
-DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
-DSHERPA_ONNX_ENABLE_JNI=ON \
..
make -j4
ls -lh lib
popd
fi
if [ ! -f ../sherpa-onnx/java-api/build/sherpa-onnx.jar ]; then
pushd ../sherpa-onnx/java-api
make
popd
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
java \
-Djava.library.path=$PWD/../build/lib \
-cp ../sherpa-onnx/java-api/build/sherpa-onnx.jar \
NonStreamingDecodeFileWhisperMultiple.java
... ...
... ... @@ -17,6 +17,14 @@ public class OfflineRecognizer {
decode(ptr, s.getPtr());
}
public void decode(OfflineStream[] ss) {
long[] streamPtrs = new long[ss.length];
for (int i = 0; i < ss.length; ++i) {
streamPtrs[i] = ss[i].getPtr();
}
decodeStreams(ptr, streamPtrs);
}
public OfflineStream createStream() {
long p = createStream(ptr);
return new OfflineStream(p);
... ... @@ -55,5 +63,7 @@ public class OfflineRecognizer {
private native void decode(long ptr, long streamPtr);
private native void decodeStreams(long ptr, long[] streamPtrs);
private native Object[] getResult(long streamPtr);
}
... ...
... ... @@ -18,6 +18,14 @@ public class OnlineRecognizer {
decode(ptr, s.getPtr());
}
public void decode(OnlineStream[] ss) {
long[] streamPtrs = new long[ss.length];
for (int i = 0; i < ss.length; ++i) {
streamPtrs[i] = ss[i].getPtr();
}
decodeStreams(ptr, streamPtrs);
}
public boolean isReady(OnlineStream s) {
return isReady(ptr, s.getPtr());
}
... ... @@ -68,6 +76,8 @@ public class OnlineRecognizer {
private native void decode(long ptr, long streamPtr);
private native void decodeStreams(long ptr, long[] streamPtrs);
private native boolean isEndpoint(long ptr, long streamPtr);
private native boolean isReady(long ptr, long streamPtr);
... ...
... ... @@ -366,22 +366,44 @@ Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_createStream(JNIEnv * /*env*/,
SHERPA_ONNX_EXTERN_C
JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_decode(
JNIEnv *env, jobject /*obj*/, jlong ptr, jlong streamPtr) {
JNIEnv *env, jobject /*obj*/, jlong ptr, jlong stream_ptr) {
SafeJNI(env, "OfflineRecognizer_decode", [&] {
if (!ValidatePointer(env, ptr, "OfflineRecognizer_decode",
"OfflineRecognizer pointer is null.") ||
!ValidatePointer(env, streamPtr, "OfflineRecognizer_decode",
!ValidatePointer(env, stream_ptr, "OfflineRecognizer_decode",
"OfflineStream pointer is null.")) {
return;
}
auto recognizer = reinterpret_cast<sherpa_onnx::OfflineRecognizer *>(ptr);
auto stream = reinterpret_cast<sherpa_onnx::OfflineStream *>(streamPtr);
auto stream = reinterpret_cast<sherpa_onnx::OfflineStream *>(stream_ptr);
recognizer->DecodeStream(stream);
});
}
SHERPA_ONNX_EXTERN_C
JNIEXPORT void JNICALL
Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_decodeStreams(
JNIEnv *env, jobject /*obj*/, jlong ptr, jlongArray stream_ptrs) {
SafeJNI(env, "OfflineRecognizer_decode_streams", [&] {
if (!ValidatePointer(env, ptr, "OfflineRecognizer_decode_streams",
"OfflineRecognizer pointer is null.")) {
return;
}
auto recognizer = reinterpret_cast<sherpa_onnx::OfflineRecognizer *>(ptr);
jlong *p = env->GetLongArrayElements(stream_ptrs, nullptr);
jsize n = env->GetArrayLength(stream_ptrs);
auto ss = reinterpret_cast<sherpa_onnx::OfflineStream **>(p);
recognizer->DecodeStreams(ss, n);
env->ReleaseLongArrayElements(stream_ptrs, p, JNI_ABORT);
});
}
SHERPA_ONNX_EXTERN_C
JNIEXPORT jobjectArray JNICALL
Java_com_k2fsa_sherpa_onnx_OfflineRecognizer_getResult(JNIEnv *env,
jobject /*obj*/,
... ...
... ... @@ -340,6 +340,22 @@ JNIEXPORT void JNICALL Java_com_k2fsa_sherpa_onnx_OnlineRecognizer_decode(
}
SHERPA_ONNX_EXTERN_C
JNIEXPORT void JNICALL
Java_com_k2fsa_sherpa_onnx_OnlineRecognizer_decodeStreams(
JNIEnv *env, jobject /*obj*/, jlong ptr, jlongArray stream_ptrs) {
auto recognizer = reinterpret_cast<sherpa_onnx::OnlineRecognizer *>(ptr);
jlong *p = env->GetLongArrayElements(stream_ptrs, nullptr);
jsize n = env->GetArrayLength(stream_ptrs);
auto ss = reinterpret_cast<sherpa_onnx::OnlineStream **>(p);
recognizer->DecodeStreams(ss, n);
env->ReleaseLongArrayElements(stream_ptrs, p, JNI_ABORT);
}
SHERPA_ONNX_EXTERN_C
JNIEXPORT jlong JNICALL
Java_com_k2fsa_sherpa_onnx_OnlineRecognizer_createStream(JNIEnv *env,
jobject /*obj*/,
... ...