Java API for speaker diarization (#1416)

java-api-examples/OfflineSpeakerDiarizationDemo.java

@@ -0,0 +1,99 @@
+// Copyright 2024 Xiaomi Corporation
+
+// This file shows how to use sherpa-onnx Java API for speaker diarization,
+import com.k2fsa.sherpa.onnx.*;
+
+public class OfflineSpeakerDiarizationDemo {
+  public static void main(String[] args) {
+    /* Please use the following commands to download files used in this file
+    Step 1: Download a speaker segmentation model
+
+    Please visit https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-segmentation-models
+    for a list of available models. The following is an example
+
+      wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+      tar xvf sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+      rm sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+
+    Step 2: Download a speaker embedding extractor model
+
+    Please visit https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-recongition-models
+    for a list of available models. The following is an example
+
+      wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+
+    Step 3. Download test wave files
+
+    Please visit https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-segmentation-models
+    for a list of available test wave files. The following is an example
+
+      wget https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/0-four-speakers-zh.wav
+
+    Step 4. Run it
+        */
+
+    String segmentationModel = "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx";
+    String embeddingModel = "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx";
+    String waveFilename = "./0-four-speakers-zh.wav";
+
+    WaveReader reader = new WaveReader(waveFilename);
+
+    OfflineSpeakerSegmentationPyannoteModelConfig pyannote =
+        OfflineSpeakerSegmentationPyannoteModelConfig.builder().setModel(segmentationModel).build();
+
+    OfflineSpeakerSegmentationModelConfig segmentation =
+        OfflineSpeakerSegmentationModelConfig.builder()
+            .setPyannote(pyannote)
+            .setDebug(true)
+            .build();
+
+    SpeakerEmbeddingExtractorConfig embedding =
+        SpeakerEmbeddingExtractorConfig.builder().setModel(embeddingModel).setDebug(true).build();
+
+    // The test wave file ./0-four-speakers-zh.wav contains four speakers, so
+    // we use numClusters=4 here. If you don't know the number of speakers
+    // in the test wave file, please set the numClusters to -1 and provide
+    // threshold for clustering
+    FastClusteringConfig clustering =
+        FastClusteringConfig.builder()
+            .setNumClusters(4) // set it to -1 if you don't know the actual number
+            .setThreshold(0.5f)
+            .build();
+
+    OfflineSpeakerDiarizationConfig config =
+        OfflineSpeakerDiarizationConfig.builder()
+            .setSegmentation(segmentation)
+            .setEmbedding(embedding)
+            .setClustering(clustering)
+            .setMinDurationOn(0.2f)
+            .setMinDurationOff(0.5f)
+            .build();
+
+    OfflineSpeakerDiarization sd = new OfflineSpeakerDiarization(config);
+    if (sd.getSampleRate() != reader.getSampleRate()) {
+      System.out.printf(
+          "Expected sample rate: %d, given: %d\n", sd.getSampleRate(), reader.getSampleRate());
+      return;
+    }
+
+    // OfflineSpeakerDiarizationSegment[] segments = sd.process(reader.getSamples());
+    // without callback is also ok
+
+    // or you can use a callback to show the progress
+    OfflineSpeakerDiarizationSegment[] segments =
+        sd.processWithCallback(
+            reader.getSamples(),
+            (int numProcessedChunks, int numTotalChunks, long arg) -> {
+              float progress = 100.0f * numProcessedChunks / numTotalChunks;
+              System.out.printf("Progress: %.2f%%\n", progress);
+
+              return 0;
+            });
+
+    for (OfflineSpeakerDiarizationSegment s : segments) {
+      System.out.printf("%.3f -- %.3f speaker_%02d\n", s.getStart(), s.getEnd(), s.getSpeaker());
+    }
+
+    sd.release();
+  }
+}

java-api-examples/README.md

@@ -4,6 +4,12 @@ This directory contains examples for the JAVA API of sherpa-onnx.
 
 # Usage
 
+## Non-streaming speaker diarization
+
+```bash
+./run-offline-speaker-diarization.sh
+```
+
 ## Streaming Speech recognition
 
 ```

java-api-examples/run-offline-speaker-diarization.sh

@@ -0,0 +1,45 @@
+#!/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-pyannote-segmentation-3-0/model.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+  tar xvf sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+  rm sherpa-onnx-pyannote-segmentation-3-0.tar.bz2
+fi
+
+if [ ! -f ./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+fi
+
+if [ ! -f ./0-four-speakers-zh.wav ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/0-four-speakers-zh.wav
+fi
+
+java \
+  -Djava.library.path=$PWD/../build/lib \
+  -cp ../sherpa-onnx/java-api/build/sherpa-onnx.jar \
+  ./OfflineSpeakerDiarizationDemo.java