Add C++ runtime for MeloTTS (#1138)

scripts/melo-tts/README.md

@@ -0,0 +1,6 @@
+# Introduction
+
+Models in this directory are converted from
+https://github.com/myshell-ai/MeloTTS
+
+Note there is only a single female speaker in the model.

scripts/melo-tts/export-onnx.py

@@ -8,7 +8,6 @@ from melo.text import language_id_map, language_tone_start_map
 from melo.text.chinese import pinyin_to_symbol_map
 from melo.text.english import eng_dict, refine_syllables
 from pypinyin import Style, lazy_pinyin, phrases_dict, pinyin_dict
-from melo.text.symbols import language_tone_start_map
 
 for k, v in pinyin_to_symbol_map.items():
     if isinstance(v, list):
@@ -82,6 +81,7 @@ def generate_tokens(symbol_list):
 def generate_lexicon():
     word_dict = pinyin_dict.pinyin_dict
     phrases = phrases_dict.phrases_dict
+    eng_dict["kaldi"] = [["K", "AH0"], ["L", "D", "IH0"]]
     with open("lexicon.txt", "w", encoding="utf-8") as f:
         for word in eng_dict:
             phones, tones = refine_syllables(eng_dict[word])
@@ -237,9 +237,11 @@ def main():
     meta_data = {
         "model_type": "melo-vits",
         "comment": "melo",
+        "version": 2,
         "language": "Chinese + English",
         "add_blank": int(model.hps.data.add_blank),
         "n_speakers": 1,
+        "jieba": 1,
         "sample_rate": model.hps.data.sampling_rate,
         "bert_dim": 1024,
         "ja_bert_dim": 768,

scripts/melo-tts/run.sh

@@ -12,7 +12,7 @@ function install() {
   cd MeloTTS
   pip install -r ./requirements.txt
 
-  pip install soundfile onnx onnxruntime
+  pip install soundfile onnx==1.15.0 onnxruntime==1.16.3
 
   python3 -m unidic download
   popd

scripts/melo-tts/test.py

@@ -135,28 +135,11 @@ class OnnxModel:
 def main():
     lexicon = Lexicon(lexion_filename="./lexicon.txt", tokens_filename="./tokens.txt")
 
-    text = "永远相信，美好的事情即将发生。"
+    text = "这是一个使用 next generation kaldi 的 text to speech 中英文例子. Thank you! 你觉得如何呢? are you ok? Fantastic! How about you?"
     s = jieba.cut(text, HMM=True)
 
     phones, tones = lexicon.convert(s)
 
-    en_text = "how are you ?".split()
-
-    phones_en, tones_en = lexicon.convert(en_text)
-    phones += [0]
-    tones += [0]
-
-    phones += phones_en
-    tones += tones_en
-
-    text = "多音字测试， 银行，行不行？长沙长大"
-    s = jieba.cut(text, HMM=True)
-
-    phones2, tones2 = lexicon.convert(s)
-
-    phones += phones2
-    tones += tones2
-
     model = OnnxModel("./model.onnx")
 
     if model.add_blank: