add more models for speaker diarization (#1440)

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

@@ -17,8 +17,9 @@ val segmentationModel = "segmentation.onnx"
 
 // please download it from
 // https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+// and rename it to embedding.onnx
 // and move it to the assets folder
-val embeddingModel = "3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx"
+val embeddingModel = "embedding.onnx"
 
 // in the end, your assets folder should look like below
 /*
@@ -26,7 +27,7 @@ val embeddingModel = "3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
 /Users/fangjun/open-source/sherpa-onnx/android/SherpaOnnxSpeakerDiarization/app/src/main/assets
 (py38) fangjuns-MacBook-Pro:assets fangjun$ ls -lh
 total 89048
--rw-r--r--  1 fangjun  staff    38M Oct 12 20:28 3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx
+-rw-r--r--  1 fangjun  staff    38M Oct 12 20:28 embedding.onnx
 -rw-r--r--  1 fangjun  staff   5.7M Oct 12 20:28 segmentation.onnx
  */
 
@@ -63,4 +64,4 @@ object SpeakerDiarizationObject {
             _sd = OfflineSpeakerDiarization(assetManager = assetManager, config = config)
         }
     }
-}
+}

scripts/apk/build-apk-speaker-diarization.sh.in

@@ -37,18 +37,20 @@ pushd ./android/SherpaOnnxSpeakerDiarization/app/src/main/assets/
 
 ls -lh
 
-model_name={{ model.model_name }}
+segmentation_model_name={{ model.segmentation.model_name }}
-short_name={{ model.short_name }}
+segmentation_short_name={{ model.segmentation.short_name }}
 
-curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-segmentation-models/$model_name.tar.bz2
+embedding_model_name={{ model.embedding.model_name }}
-tar xvf $model_name.tar.bz2
+embedding_short_name={{ model.embedding.short_name }}
-rm $model_name.tar.bz2
-mv $model_name/model.onnx segmentation.onnx
-rm -rf $model_name
 
-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-segmentation-models/$segmentation_model_name.tar.bz2
-  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
+tar xvf $segmentation_model_name.tar.bz2
-fi
+rm $segmentation_model_name.tar.bz2
+mv $segmentation_model_name/model.onnx segmentation.onnx
+rm -rf $segmentation_model_name
+
+curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/$embedding_model_name.onnx
+mv $embedding_model_name.onnx embedding.onnx
 
 echo "pwd: $PWD"
 ls -lh
@@ -74,12 +76,12 @@ for arch in arm64-v8a armeabi-v7a x86_64 x86; do
   ./gradlew build
   popd
 
-  mv android/SherpaOnnxSpeakerDiarization/app/build/outputs/apk/debug/app-debug.apk ./apks/sherpa-onnx-${SHERPA_ONNX_VERSION}-$arch-speaker-diarization-$short_name-3dspeaker.apk
+  mv android/SherpaOnnxSpeakerDiarization/app/build/outputs/apk/debug/app-debug.apk ./apks/sherpa-onnx-${SHERPA_ONNX_VERSION}-$arch-speaker-diarization-$segmentation_short_name-$embedding_short_name.apk
   ls -lh apks
   rm -v ./android/SherpaOnnxSpeakerDiarization/app/src/main/jniLibs/$arch/*.so
 done
 
-rm -rf ./android/SherpaOnnxSpeakerDiarization/app/src/main/assets/segmentation.onnx
+rm -rf ./android/SherpaOnnxSpeakerDiarization/app/src/main/assets/*.onnx
 
 {% endfor %}
 

scripts/apk/generate-speaker-diarization-apk-script.py

@@ -27,10 +27,22 @@ def get_args():
 @dataclass
 class SpeakerSegmentationModel:
     model_name: str
-    short_name: str = ""
+    short_name: str
 
 
-def get_models() -> List[SpeakerSegmentationModel]:
+@dataclass
+class SpeakerEmbeddingModel:
+    model_name: str
+    short_name: str
+
+
+@dataclass
+class Model:
+    segmentation: SpeakerSegmentationModel
+    embedding: SpeakerEmbeddingModel
+
+
+def get_segmentation_models() -> List[SpeakerSegmentationModel]:
     models = [
         SpeakerSegmentationModel(
             model_name="sherpa-onnx-pyannote-segmentation-3-0",
@@ -45,13 +57,33 @@ def get_models() -> List[SpeakerSegmentationModel]:
     return models
 
 
+def get_embedding_models() -> List[SpeakerEmbeddingModel]:
+    models = [
+        SpeakerSegmentationModel(
+            model_name="3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k",
+            short_name="3dspeaker",
+        ),
+        SpeakerSegmentationModel(
+            model_name="nemo_en_titanet_small",
+            short_name="nemo",
+        ),
+    ]
+    return models
+
+
 def main():
     args = get_args()
     index = args.index
     total = args.total
     assert 0 <= index < total, (index, total)
 
-    all_model_list = get_models()
+    segmentation_models = get_segmentation_models()
+    embedding_models = get_embedding_models()
+
+    all_model_list = []
+    for s in segmentation_models:
+        for e in embedding_models:
+            all_model_list.append(Model(segmentation=s, embedding=e))
 
     num_models = len(all_model_list)