Add Android demo for speaker recognition (#536)

See pre-built Android APKs at 
https://k2-fsa.github.io/sherpa/onnx/speaker-identification/apk.html

scripts/apk/.gitignore

@@ -1 +1,2 @@
 build-apk-tts.sh
+!*.sh.in

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

@@ -0,0 +1,81 @@
+#!/usr/bin/env bash
+#
+# Auto generated! Please DO NOT EDIT!
+
+# Please set the environment variable ANDROID_NDK
+# before running this script
+
+# Inside the $ANDROID_NDK directory, you can find a binary ndk-build
+# and some other files like the file "build/cmake/android.toolchain.cmake"
+
+set -ex
+
+log() {
+  # This function is from espnet
+  local fname=${BASH_SOURCE[1]##*/}
+  echo -e "$(date '+%Y-%m-%d %H:%M:%S') (${fname}:${BASH_LINENO[0]}:${FUNCNAME[1]}) $*"
+}
+
+SHERPA_ONNX_VERSION=$(grep "SHERPA_ONNX_VERSION" ./CMakeLists.txt  | cut -d " " -f 2  | cut -d '"' -f 2)
+
+log "Building Speaker identification APK for sherpa-onnx v${SHERPA_ONNX_VERSION}"
+
+log "====================arm64-v8a================="
+./build-android-arm64-v8a.sh
+log "====================armv7-eabi================"
+./build-android-armv7-eabi.sh
+log "====================x86-64===================="
+./build-android-x86-64.sh
+log "====================x86===================="
+./build-android-x86.sh
+
+mkdir -p apks
+
+{% for model in model_list %}
+pushd ./android/SherpaOnnxSpeakerIdentification/app/src/main/assets/
+model_name={{ model.model_name }}
+short_name={{ model.short_name }}
+lang={{ model.lang }}
+framework={{ model.framework }}
+
+wget -qq https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/$model_name
+
+popd
+# Now we are at the project root directory
+
+git checkout .
+pushd android/SherpaOnnxSpeakerIdentification/app/src/main/java/com/k2fsa/sherpa/onnx/speaker/identification/
+sed -i.bak s/"private val modelName.*/private val modelName = \"$model_name\"/" ./Speaker.kt
+git diff
+popd
+
+for arch in arm64-v8a armeabi-v7a x86_64 x86; do
+  log "------------------------------------------------------------"
+  log "build tts apk for $arch"
+  log "------------------------------------------------------------"
+  src_arch=$arch
+  if [ $arch == "armeabi-v7a" ]; then
+    src_arch=armv7-eabi
+  elif [ $arch == "x86_64" ]; then
+    src_arch=x86-64
+  fi
+
+  ls -lh ./build-android-$src_arch/install/lib/*.so
+
+  cp -v ./build-android-$src_arch/install/lib/*.so ./android/SherpaOnnxSpeakerIdentification/app/src/main/jniLibs/$arch/
+
+  pushd ./android/SherpaOnnxSpeakerIdentification
+  ./gradlew build
+  popd
+
+  mv android/SherpaOnnxSpeakerIdentification/app/build/outputs/apk/debug/app-debug.apk ./apks/sherpa-onnx-${SHERPA_ONNX_VERSION}-$arch-$lang-speaker-identification-$framework-$short_name.apk
+  ls -lh apks
+  rm -v ./android/SherpaOnnxSpeakerIdentification/app/src/main/jniLibs/$arch/*.so
+done
+
+rm -rf ./android/SherpaOnnxSpeakerIdentification/app/src/main/assets/$model_name
+{% endfor %}
+
+git checkout .
+
+ls -lh apks/

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

@@ -0,0 +1,150 @@
+#!/usr/bin/env python3
+
+import argparse
+from dataclasses import dataclass
+from typing import List, Optional
+
+import jinja2
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--total",
+        type=int,
+        default=1,
+        help="Number of runners",
+    )
+    parser.add_argument(
+        "--index",
+        type=int,
+        default=0,
+        help="Index of the current runner",
+    )
+    return parser.parse_args()
+
+
+@dataclass
+class SpeakerIdentificationModel:
+    model_name: str
+    short_name: str = ""
+    lang: str = ""
+    framework: str = ""
+
+
+def get_3dspeaker_models() -> List[SpeakerIdentificationModel]:
+    models = [
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_campplus_sv_en_voxceleb_16k.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_campplus_sv_zh-cn_16k-common.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_eres2net_base_200k_sv_zh-cn_16k-common.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_eres2net_large_sv_zh-cn_3dspeaker_16k.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_eres2net_sv_en_voxceleb_16k.onnx"),
+        SpeakerIdentificationModel(model_name="3dspeaker_speech_eres2net_sv_zh-cn_16k-common.onnx"),
+    ]
+
+    prefix = '3dspeaker_speech_'
+    num = len(prefix)
+    for m in models:
+        m.framework = '3dspeaker'
+        m.short_name = m.model_name[num:-5]
+        if '_zh-cn_' in m.model_name:
+            m.lang = 'zh'
+        elif '_en_' in m.model_name:
+            m.lang = 'en'
+        else:
+            raise ValueError(m)
+    return models
+
+def get_wespeaker_models() -> List[SpeakerIdentificationModel]:
+    models = [
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_CAM++.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_CAM++_LM.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_resnet152_LM.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_resnet221_LM.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_resnet293_LM.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_resnet34.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_en_voxceleb_resnet34_LM.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_zh_cnceleb_resnet34.onnx"),
+        SpeakerIdentificationModel(model_name="wespeaker_zh_cnceleb_resnet34_LM.onnx"),
+    ]
+
+    prefix = 'wespeaker_xx_'
+    num = len(prefix)
+    for m in models:
+        m.framework = 'wespeaker'
+        m.short_name = m.model_name[num:-5]
+        if '_zh_' in m.model_name:
+            m.lang = 'zh'
+        elif '_en_' in m.model_name:
+            m.lang = 'en'
+        else:
+            raise ValueError(m)
+    return models
+
+def get_nemo_models() -> List[SpeakerIdentificationModel]:
+    models = [
+        SpeakerIdentificationModel(model_name="nemo_en_speakerverification_speakernet.onnx"),
+        SpeakerIdentificationModel(model_name="nemo_en_titanet_large.onnx"),
+        SpeakerIdentificationModel(model_name="nemo_en_titanet_small.onnx"),
+    ]
+
+    prefix = 'nemo_en_'
+    num = len(prefix)
+    for m in models:
+        m.framework = 'nemo'
+        m.short_name = m.model_name[num:-5]
+        if '_zh_' in m.model_name:
+            m.lang = 'zh'
+        elif '_en_' in m.model_name:
+            m.lang = 'en'
+        else:
+            raise ValueError(m)
+    return models
+
+
+
+def main():
+    args = get_args()
+    index = args.index
+    total = args.total
+    assert 0 <= index < total, (index, total)
+
+    all_model_list = get_3dspeaker_models()
+    all_model_list += get_wespeaker_models()
+    all_model_list += get_nemo_models()
+
+    num_models = len(all_model_list)
+
+    num_per_runner = num_models // total
+    if num_per_runner <= 0:
+        raise ValueError(f"num_models: {num_models}, num_runners: {total}")
+
+    start = index * num_per_runner
+    end = start + num_per_runner
+
+    remaining = num_models - args.total * num_per_runner
+
+    print(f"{index}/{total}: {start}-{end}/{num_models}")
+
+    d = dict()
+    d["model_list"] = all_model_list[start:end]
+    if index < remaining:
+        s = args.total * num_per_runner + index
+        d["model_list"].append(all_model_list[s])
+        print(f"{s}/{num_models}")
+
+    filename_list = ["./build-apk-speaker-identification.sh"]
+    for filename in filename_list:
+        environment = jinja2.Environment()
+        with open(f"{filename}.in") as f:
+            s = f.read()
+        template = environment.from_string(s)
+
+        s = template.render(**d)
+        with open(filename, "w") as f:
+            print(s, file=f)
+
+
+if __name__ == "__main__":
+    main()