Expose the pyannote window shift ratio in the C API and language bindings

android/SherpaOnnxSpeakerDiarization/app/src/main/java/com/k2fsa/sherpa/onnx/speaker/diarization/SpeakerDiarizationObject.kt

@@ -48,7 +48,8 @@ object SpeakerDiarizationObject {
             val config = OfflineSpeakerDiarizationConfig(
                 segmentation = OfflineSpeakerSegmentationModelConfig(
                     pyannote = OfflineSpeakerSegmentationPyannoteModelConfig(
-                        segmentationModel
+                        model = segmentationModel,
+                        windowShiftRatio = 0.1f,
                     ),
                     debug = true,
                 ),

c-api-examples/offline-speaker-diarization-c-api.c

@@ -73,6 +73,8 @@ int main() {
   memset(&config, 0, sizeof(config));
 
   config.segmentation.pyannote.model = segmentation_model;
+  // A value of 0 also uses the default of 0.1.
+  config.segmentation.pyannote.window_shift_ratio = 0.1f;
   config.embedding.model = embedding_extractor_model;
 
   // the test wave ./0-four-speakers-zh.wav has 4 speakers, so

cxx-api-examples/offline-speaker-diarization-cxx-api.cc

@@ -31,6 +31,7 @@ int32_t main() {
   OfflineSpeakerDiarizationConfig config;
   config.segmentation.pyannote.model =
       "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx";
+  config.segmentation.pyannote.window_shift_ratio = 0.1f;
   config.embedding.model =
       "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx";
   config.clustering.num_clusters = 4;

dart-api-examples/speaker-diarization/bin/speaker-diarization.dart

@@ -46,7 +46,7 @@ void main(List<String> arguments) async {
 
   final segmentationConfig = sherpa_onnx.OfflineSpeakerSegmentationModelConfig(
     pyannote: sherpa_onnx.OfflineSpeakerSegmentationPyannoteModelConfig(
-        model: segmentationModel),
+        model: segmentationModel, windowShiftRatio: 0.1),
   );
 
   final embeddingConfig =

dotnet-examples/offline-speaker-diarization/Program.cs

@@ -39,9 +39,10 @@ class OfflineSpeakerDiarizationDemo
 {
   static void Main(string[] args)
   {
-    var config = new OfflineSpeakerDiarizationConfig();
-    config.Segmentation.Pyannote.Model = "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx";
-    config.Embedding.Model = "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx";
+    var config = new OfflineSpeakerDiarizationConfig();
+    config.Segmentation.Pyannote.Model = "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx";
+    config.Segmentation.Pyannote.WindowShiftRatio = 0.1f;
+    config.Embedding.Model = "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx";
 
     // the test wave ./0-four-speakers-zh.wav has 4 speakers, so
     // we set num_clusters to 4

flutter/sherpa_onnx/lib/src/offline_speaker_diarization.dart

@@ -40,6 +40,8 @@ class OfflineSpeakerDiarization {
 
     c.ref.segmentation.pyannote.model =
         config.segmentation.pyannote.model.toNativeUtf8();
+    c.ref.segmentation.pyannote.windowShiftRatio =
+        config.segmentation.pyannote.windowShiftRatio;
     c.ref.segmentation.numThreads = config.segmentation.numThreads;
     c.ref.segmentation.debug = config.segmentation.debug ? 1 : 0;
     c.ref.segmentation.provider = config.segmentation.provider.toNativeUtf8();

flutter/sherpa_onnx/lib/src/offline_speaker_diarization_config.dart

@@ -43,25 +43,31 @@ class OfflineSpeakerDiarizationSegment {
 class OfflineSpeakerSegmentationPyannoteModelConfig {
   const OfflineSpeakerSegmentationPyannoteModelConfig({
     this.model = '',
+    this.windowShiftRatio = 0.1,
   });
 
   factory OfflineSpeakerSegmentationPyannoteModelConfig.fromJson(
       Map<String, dynamic> json) {
     return OfflineSpeakerSegmentationPyannoteModelConfig(
       model: json['model'] as String? ?? '',
+      windowShiftRatio:
+          (json['windowShiftRatio'] as num?)?.toDouble() ?? 0.1,
     );
   }
 
   @override
   String toString() {
-    return 'OfflineSpeakerSegmentationPyannoteModelConfig(model: $model)';
+    return 'OfflineSpeakerSegmentationPyannoteModelConfig(model: $model, '
+        'windowShiftRatio: $windowShiftRatio)';
   }
 
   Map<String, dynamic> toJson() => {
         'model': model,
+        'windowShiftRatio': windowShiftRatio,
       };
 
   final String model;
+  final double windowShiftRatio;
 }
 
 /// Segmentation model configuration for speaker diarization.

flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart

@@ -71,6 +71,9 @@ final class SherpaOnnxOfflineSpeakerDiarizationSegment extends Struct {
 final class SherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig
     extends Struct {
   external Pointer<Utf8> model;
+
+  @Float()
+  external double windowShiftRatio;
 }
 
 final class SherpaOnnxOfflineSpeakerSegmentationModelConfig extends Struct {

go-api-examples/non-streaming-speaker-diarization/main.go

@@ -38,6 +38,7 @@ func initSpeakerDiarization() *sherpa.OfflineSpeakerDiarization {
 	config := sherpa.OfflineSpeakerDiarizationConfig{}
 
 	config.Segmentation.Pyannote.Model = "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx"
+	config.Segmentation.Pyannote.WindowShiftRatio = 0.1
 	config.Embedding.Model = "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx"
 
 	// The test wave file contains 4 speakers, so we use 4 here

harmony-os/SherpaOnnxHar/sherpa_onnx/src/main/cpp/non-streaming-speaker-diarization.cc

@@ -23,6 +23,7 @@ GetOfflineSpeakerSegmentationPyannoteModelConfig(Napi::Object obj) {
 
   Napi::Object o = obj.Get("pyannote").As<Napi::Object>();
   SHERPA_ONNX_ASSIGN_ATTR_STR(model, model);
+  SHERPA_ONNX_ASSIGN_ATTR_FLOAT(window_shift_ratio, windowShiftRatio);
 
   return c;
 }

harmony-os/SherpaOnnxHar/sherpa_onnx/src/main/ets/components/NonStreamingSpeakerDiarization.ets

@@ -10,6 +10,7 @@ import { SpeakerEmbeddingExtractorConfig } from './SpeakerIdentification';
 
 export class OfflineSpeakerSegmentationPyannoteModelConfig {
   public model: string = '';
+  public windowShiftRatio: number = 0.1;
 }
 
 export class OfflineSpeakerSegmentationModelConfig {

harmony-os/SherpaOnnxSpeakerDiarization/entry/src/main/ets/workers/SpeakerDiarizationWorker.ets

@@ -32,6 +32,7 @@ function initOfflineSpeakerDiarization(context: Context): OfflineSpeakerDiarizat
   //
   // Also, please delete unused files to reduce the size of the app
   config.segmentation.pyannote.model = 'sherpa-onnx-pyannote-segmentation-3-0/model.int8.onnx';
+  config.segmentation.pyannote.windowShiftRatio = 0.1;
   config.segmentation.numThreads = 2;
   config.segmentation.debug = true;
 
@@ -186,4 +187,4 @@ workerPort.onmessageerror = (e: MessageEvents) => {
  * @param e error message
  */
 workerPort.onerror = (e: ErrorEvent) => {
-}
+}

java-api-examples/OfflineSpeakerDiarizationDemo.java

@@ -39,7 +39,10 @@ public static void main(String[] args) {
     WaveReader reader = new WaveReader(waveFilename);
 
     OfflineSpeakerSegmentationPyannoteModelConfig pyannote =
-        OfflineSpeakerSegmentationPyannoteModelConfig.builder().setModel(segmentationModel).build();
+        OfflineSpeakerSegmentationPyannoteModelConfig.builder()
+            .setModel(segmentationModel)
+            .setWindowShiftRatio(0.1f)
+            .build();
 
     OfflineSpeakerSegmentationModelConfig segmentation =
         OfflineSpeakerSegmentationModelConfig.builder()

kotlin-api-examples/test_offline_speaker_diarization.kt

@@ -15,7 +15,10 @@ fun callback(numProcessedChunks: Int, numTotalChunks: Int, arg: Long): Int {
 fun testOfflineSpeakerDiarization() {
   var config = OfflineSpeakerDiarizationConfig(
     segmentation=OfflineSpeakerSegmentationModelConfig(
-      pyannote=OfflineSpeakerSegmentationPyannoteModelConfig("./sherpa-onnx-pyannote-segmentation-3-0/model.onnx"),
+      pyannote=OfflineSpeakerSegmentationPyannoteModelConfig(
+        model="./sherpa-onnx-pyannote-segmentation-3-0/model.onnx",
+        windowShiftRatio=0.1f,
+      ),
     ),
     embedding=SpeakerEmbeddingExtractorConfig(
       model="./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx",

nodejs-addon-examples/test_offline_speaker_diarization.js

@@ -20,6 +20,7 @@ const config = {
   segmentation: {
     pyannote: {
       model: './sherpa-onnx-pyannote-segmentation-3-0/model.onnx',
+      windowShiftRatio: 0.1,
     },
   },
   embedding: {

nodejs-examples/test-offline-speaker-diarization.js

@@ -20,6 +20,7 @@ const config = {
   segmentation: {
     pyannote: {
       model: './sherpa-onnx-pyannote-segmentation-3-0/model.onnx',
+      windowShiftRatio: 0.1,
       debug: 1,
     },
   },

pascal-api-examples/speaker-diarization/main.pas

@@ -62,6 +62,7 @@ function ProgressCallback(
   Wave := SherpaOnnxReadWave('./0-four-speakers-zh.wav');
 
   Config.Segmentation.Pyannote.Model := './sherpa-onnx-pyannote-segmentation-3-0/model.onnx';
+  Config.Segmentation.Pyannote.WindowShiftRatio := 0.1;
   Config.Embedding.Model := './3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx';
 
   {

python-api-examples/offline-speaker-diarization.py

@@ -73,7 +73,7 @@ def init_speaker_diarization(num_speakers: int = -1, cluster_threshold: float =
     config = sherpa_onnx.OfflineSpeakerDiarizationConfig(
         segmentation=sherpa_onnx.OfflineSpeakerSegmentationModelConfig(
             pyannote=sherpa_onnx.OfflineSpeakerSegmentationPyannoteModelConfig(
-                model=segmentation_model
+                model=segmentation_model, window_shift_ratio=0.1
             ),
         ),
         embedding=sherpa_onnx.SpeakerEmbeddingExtractorConfig(

rust-api-examples/examples/offline_speaker_diarization.rs

@@ -9,6 +9,7 @@ fn main() {
         segmentation: OfflineSpeakerSegmentationModelConfig {
             pyannote: OfflineSpeakerSegmentationPyannoteModelConfig {
                 model: Some("./sherpa-onnx-pyannote-segmentation-3-0/model.onnx".into()),
+                window_shift_ratio: 0.1,
             },
             ..Default::default()
         },

scripts/dotnet/OfflineSpeakerSegmentationPyannoteModelConfig.cs

@@ -11,10 +11,11 @@ public struct OfflineSpeakerSegmentationPyannoteModelConfig
         public OfflineSpeakerSegmentationPyannoteModelConfig()
         {
             Model = "";
+            WindowShiftRatio = 0.1f;
         }
 
         [MarshalAs(UnmanagedType.LPStr)]
         public string Model;
+        public float WindowShiftRatio;
     }
 }
-

scripts/go/sherpa_onnx.go

@@ -2080,7 +2080,8 @@ func ReadWave(filename string) *Wave {
 // For offline speaker diarization
 // ============================================================
 type OfflineSpeakerSegmentationPyannoteModelConfig struct {
-	Model string
+	Model            string
+	WindowShiftRatio float32
 }
 
 type OfflineSpeakerSegmentationModelConfig struct {
@@ -2116,6 +2117,7 @@ func NewOfflineSpeakerDiarization(config *OfflineSpeakerDiarizationConfig) *Offl
 	c := C.struct_SherpaOnnxOfflineSpeakerDiarizationConfig{}
 	c.segmentation.pyannote.model = C.CString(config.Segmentation.Pyannote.Model)
 	defer C.free(unsafe.Pointer(c.segmentation.pyannote.model))
+	c.segmentation.pyannote.window_shift_ratio = C.float(config.Segmentation.Pyannote.WindowShiftRatio)
 
 	c.segmentation.num_threads = C.int(config.Segmentation.NumThreads)
 

scripts/node-addon-api/lib/types.js

@@ -645,10 +645,11 @@
  */
 
 /**
- * Offline speaker segmentation (pyannote) model config
- * @typedef {Object} OfflineSpeakerSegmentationPyannoteModelConfig
- * @property {string} [model]
- */
+ * Offline speaker segmentation (pyannote) model config
+ * @typedef {Object} OfflineSpeakerSegmentationPyannoteModelConfig
+ * @property {string} [model]
+ * @property {number} [windowShiftRatio=0.1]
+ */
 
 /**
  * Offline speaker segmentation model config

sherpa-onnx/c-api/c-api.cc

@@ -3149,6 +3149,10 @@ GetOfflineSpeakerDiarizationConfig(
 
   sd_config.segmentation.pyannote.model =
       SHERPA_ONNX_OR(config->segmentation.pyannote.model, "");
+  sd_config.segmentation.pyannote.window_shift_ratio =
+      config->segmentation.pyannote.window_shift_ratio <= 0
+          ? 0.1f
+          : config->segmentation.pyannote.window_shift_ratio;
   sd_config.segmentation.num_threads =
       GetNumThreads(config->segmentation.num_threads);
   sd_config.segmentation.debug = config->segmentation.debug;

sherpa-onnx/c-api/c-api.h

@@ -3836,6 +3836,13 @@ SHERPA_ONNX_API int32_t SherpaOnnxLinearResamplerResampleGetOutputSampleRate(
 typedef struct SherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig {
   /** Segmentation model filename. */
   const char *model;
+  /**
+   * Sliding-window shift as a fraction of the model window size.
+   *
+   * Must be in (0, 1]. Set to 0 (or a negative value) to use the default of
+   * 0.1.
+   */
+  float window_shift_ratio;
 } SherpaOnnxOfflineSpeakerSegmentationPyannoteModelConfig;
 
 /**

sherpa-onnx/c-api/cxx-api.cc

@@ -1464,6 +1464,8 @@ OfflineSpeakerDiarization OfflineSpeakerDiarization::Create(
   memset(&c, 0, sizeof(c));
   c.segmentation.pyannote.model =
       config.segmentation.pyannote.model.c_str();
+  c.segmentation.pyannote.window_shift_ratio =
+      config.segmentation.pyannote.window_shift_ratio;
   c.segmentation.num_threads = config.segmentation.num_threads;
   c.segmentation.debug = config.segmentation.debug;
   c.segmentation.provider = config.segmentation.provider.c_str();

sherpa-onnx/c-api/cxx-api.h

@@ -1912,6 +1912,8 @@ class SHERPA_ONNX_API SpeakerEmbeddingManager
 struct OfflineSpeakerSegmentationPyannoteModelConfig {
   /** Segmentation model filename. */
   std::string model;
+  /** Sliding-window shift as a fraction of the model window size. */
+  float window_shift_ratio = 0.1f;
 };
 
 /** @brief Segmentation model configuration for offline speaker diarization. */

sherpa-onnx/c-api/docs/speaker-diarization.dox

@@ -14,6 +14,10 @@ SherpaOnnxOfflineSpeakerDiarizationConfig config;
 memset(&config, 0, sizeof(config));
 config.segmentation.pyannote.model =
     "./sherpa-onnx-pyannote-segmentation-3-0/model.onnx";
+// Zero uses the default window shift ratio of 0.1. Set a value in (0, 1] to
+// change the sliding-window shift; smaller values mean more overlap and more
+// compute.
+config.segmentation.pyannote.window_shift_ratio = 0.1f;
 config.embedding.model =
     "./3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx";
 config.clustering.num_clusters = 4;

sherpa-onnx/csrc/offline-speaker-segmentation-pyannote-model.cc

@@ -112,6 +112,11 @@ class OfflineSpeakerSegmentationPyannoteModel::Impl {
       meta_data_.window_shift = static_cast<int32_t>(window_shift);
     }
 
+    if (config_.debug) {
+      SHERPA_ONNX_LOGE("Pyannote window shift: %d samples",
+                       meta_data_.window_shift);
+    }
+
     SHERPA_ONNX_READ_META_DATA(meta_data_.receptive_field_size,
                                "receptive_field_size");
     SHERPA_ONNX_READ_META_DATA(meta_data_.receptive_field_shift,

sherpa-onnx/java-api/src/main/java/com/k2fsa/sherpa/onnx/OfflineSpeakerSegmentationPyannoteModelConfig.java

@@ -4,9 +4,11 @@
 
 public class OfflineSpeakerSegmentationPyannoteModelConfig {
     private final String model;
+    private final float windowShiftRatio;
 
     private OfflineSpeakerSegmentationPyannoteModelConfig(Builder builder) {
         this.model = builder.model;
+        this.windowShiftRatio = builder.windowShiftRatio;
     }
 
     public static Builder builder() {
@@ -17,8 +19,13 @@ public String getModel() {
         return model;
     }
 
+    public float getWindowShiftRatio() {
+        return windowShiftRatio;
+    }
+
     public static class Builder {
         private String model = "";
+        private float windowShiftRatio = 0.1f;
 
         public OfflineSpeakerSegmentationPyannoteModelConfig build() {
             return new OfflineSpeakerSegmentationPyannoteModelConfig(this);
@@ -28,5 +35,10 @@ public Builder setModel(String model) {
             this.model = model;
             return this;
         }
+
+        public Builder setWindowShiftRatio(float value) {
+            this.windowShiftRatio = value;
+            return this;
+        }
     }
 }

sherpa-onnx/jni/offline-speaker-diarization.cc

@@ -33,6 +33,9 @@ static OfflineSpeakerDiarizationConfig GetOfflineSpeakerDiarizationConfig(
 
   SHERPA_ONNX_JNI_READ_STRING(ans.segmentation.pyannote.model, model,
                               pyannote_config_cls, pyannote_config);
+  SHERPA_ONNX_JNI_READ_FLOAT(ans.segmentation.pyannote.window_shift_ratio,
+                             windowShiftRatio, pyannote_config_cls,
+                             pyannote_config);
 
   SHERPA_ONNX_JNI_READ_INT(ans.segmentation.num_threads, numThreads,
                            segmentation_config_cls, segmentation_config);

sherpa-onnx/kotlin-api/OfflineSpeakerDiarization.kt

@@ -4,6 +4,7 @@ import android.content.res.AssetManager
 
 data class OfflineSpeakerSegmentationPyannoteModelConfig(
     var model: String = "",
+    var windowShiftRatio: Float = 0.1f,
 )
 
 data class OfflineSpeakerSegmentationModelConfig(