Add Android demo for MatchaTTS models. (#1683)

scripts/apk/build-apk-tts-engine.sh.in

@@ -37,6 +37,8 @@ mkdir -p apks
 pushd ./android/SherpaOnnxTtsEngine/app/src/main/assets/
 model_dir={{ tts_model.model_dir }}
 model_name={{ tts_model.model_name }}
+acoustic_model_name={{ tts_model.acoustic_model_name }}
+vocoder={{ tts_model.vocoder }}
 lang={{ tts_model.lang }}
 lang_iso_639_3={{ tts_model.lang_iso_639_3 }}
 
@@ -44,15 +46,30 @@ wget -qq https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/$mod
 tar xf $model_dir.tar.bz2
 rm $model_dir.tar.bz2
 
+{% if tts_model.vocoder %}
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/$vocoder
+{% endif %}
+
 popd
 # Now we are at the project root directory
 
 git checkout .
 pushd android/SherpaOnnxTtsEngine/app/src/main/java/com/k2fsa/sherpa/onnx/tts/engine
 sed -i.bak s/"modelDir = null"/"modelDir = \"$model_dir\""/ ./TtsEngine.kt
-sed -i.bak s/"modelName = null"/"modelName = \"$model_name\""/ ./TtsEngine.kt
 sed -i.bak s/"lang = null"/"lang = \"$lang_iso_639_3\""/ ./TtsEngine.kt
 
+{% if tts_model.model_name %}
+  sed -i.bak s/"modelName = null"/"modelName = \"$model_name\""/ ./TtsEngine.kt
+{% endif %}
+
+{% if tts_model.model_name %}
+  sed -i.bak s/"acousticModelName = null"/"acousticModelName = \"$acoustic_model_name\""/ ./TtsEngine.kt
+{% endif %}
+
+{% if tts_model.vocoder %}
+  sed -i.bak s/"vocoder = null"/"vocoder = \"$vocoder\""/ ./TtsEngine.kt
+{% endif %}
+
 {% if tts_model.rule_fsts %}
   rule_fsts={{ tts_model.rule_fsts }}
   sed -i.bak s%"ruleFsts = null"%"ruleFsts = \"$rule_fsts\""% ./TtsEngine.kt
@@ -109,6 +126,7 @@ for arch in arm64-v8a armeabi-v7a x86_64 x86; do
 done
 
 rm -rf ./android/SherpaOnnxTtsEngine/app/src/main/assets/$model_dir
+rm -fv ./android/SherpaOnnxTtsEngine/app/src/main/assets/*.onnx
 {% endfor %}
 
 git checkout .

scripts/apk/build-apk-tts.sh.in

@@ -37,19 +37,38 @@ mkdir -p apks
 pushd ./android/SherpaOnnxTts/app/src/main/assets/
 model_dir={{ tts_model.model_dir }}
 model_name={{ tts_model.model_name }}
+acoustic_model_name={{ tts_model.acoustic_model_name }}
+vocoder={{ tts_model.vocoder }}
 lang={{ tts_model.lang }}
 
 wget -qq https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/$model_dir.tar.bz2
 tar xf $model_dir.tar.bz2
 rm $model_dir.tar.bz2
 
+{% if tts_model.vocoder %}
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/$vocoder
+{% endif %}
+
 popd
 # Now we are at the project root directory
 
 git checkout .
 pushd android/SherpaOnnxTts/app/src/main/java/com/k2fsa/sherpa/onnx
 sed -i.bak s/"modelDir = null"/"modelDir = \"$model_dir\""/ ./MainActivity.kt
-sed -i.bak s/"modelName = null"/"modelName = \"$model_name\""/ ./MainActivity.kt
+
+
+{% if tts_model.model_name %}
+  sed -i.bak s/"modelName = null"/"modelName = \"$model_name\""/ ./MainActivity.kt
+{% endif %}
+
+{% if tts_model.acoustic_model_name %}
+  sed -i.bak s/"acousticModelName = null"/"acousticModelName = \"$acoustic_model_name\""/ ./MainActivity.kt
+{% endif %}
+
+{% if tts_model.vocoder %}
+  sed -i.bak s/"vocoder = null"/"vocoder = \"$vocoder\""/ ./MainActivity.kt
+{% endif %}
+
 
 {% if tts_model.rule_fsts %}
   rule_fsts={{ tts_model.rule_fsts }}
@@ -107,6 +126,8 @@ for arch in arm64-v8a armeabi-v7a x86_64 x86; do
 done
 
 rm -rf ./android/SherpaOnnxTts/app/src/main/assets/$model_dir
+rm -fv ./android/SherpaOnnxTts/app/src/main/assets/*.onnx
+
 {% endfor %}
 
 git checkout .

scripts/apk/generate-tts-apk-script.py

@@ -30,7 +30,9 @@ def get_args():
 @dataclass
 class TtsModel:
     model_dir: str
-    model_name: str = ""
+    model_name: str = ""  # for vits
+    acoustic_model_name: str = ""  # for matcha
+    vocoder: str = ""  # for matcha
     lang: str = ""  # en, zh, fr, de, etc.
     rule_fsts: Optional[List[str]] = None
     rule_fars: Optional[List[str]] = None
@@ -378,6 +380,35 @@ def get_vits_models() -> List[TtsModel]:
     return all_models
 
 
+def get_matcha_models() -> List[TtsModel]:
+    chinese_models = [
+        TtsModel(
+            model_dir="matcha-icefall-zh-baker",
+            acoustic_model_name="model-steps-3.onnx",
+            lang="zh",
+        )
+    ]
+    rule_fsts = ["phone.fst", "date.fst", "number.fst"]
+    for m in chinese_models:
+        s = [f"{m.model_dir}/{r}" for r in rule_fsts]
+        m.rule_fsts = ",".join(s)
+        m.dict_dir = m.model_dir + "/dict"
+        m.vocoder = "hifigan_v2.onnx"
+
+    english_models = [
+        TtsModel(
+            model_dir="matcha-icefall-en_US-ljspeech",
+            acoustic_model_name="model-steps-3.onnx",
+            lang="en",
+        )
+    ]
+    for m in english_models:
+        m.data_dir = f"{m.model_dir}/espeak-ng-data"
+        m.vocoder = "hifigan_v2.onnx"
+
+    return chinese_models + english_models
+
+
 def main():
     args = get_args()
     index = args.index
@@ -389,7 +420,10 @@ def main():
     all_model_list += get_piper_models()
     all_model_list += get_mimic3_models()
     all_model_list += get_coqui_models()
+    all_model_list += get_matcha_models()
+
     convert_lang_to_iso_639_3(all_model_list)
+    print(all_model_list)
 
     num_models = len(all_model_list)