Support English for MeloTTS models. (#1134)

.github/workflows/windows-x64-jni.yaml

@@ -20,7 +20,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [windows-latest]
+        os: [windows-2019]
 
     steps:
       - uses: actions/checkout@v4

scripts/melo-tts/export-onnx.py

@@ -6,9 +6,13 @@ import torch
 from melo.api import TTS
 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):
+        break
     pinyin_to_symbol_map[k] = v.split()
 
 
@@ -79,6 +83,16 @@ def generate_lexicon():
     word_dict = pinyin_dict.pinyin_dict
     phrases = phrases_dict.phrases_dict
     with open("lexicon.txt", "w", encoding="utf-8") as f:
+        for word in eng_dict:
+            phones, tones = refine_syllables(eng_dict[word])
+            tones = [t + language_tone_start_map["EN"] for t in tones]
+            tones = [str(t) for t in tones]
+
+            phones = " ".join(phones)
+            tones = " ".join(tones)
+
+            f.write(f"{word.lower()} {phones} {tones}\n")
+
         for key in word_dict:
             if not (0x4E00 <= key <= 0x9FA5):
                 continue
@@ -125,15 +139,13 @@ class ModelWrapper(torch.nn.Module):
     def __init__(self, model: "SynthesizerTrn"):
         super().__init__()
         self.model = model
+        self.lang_id = language_id_map[model.language]
 
     def forward(
         self,
         x,
         x_lengths,
         tones,
-        lang_id,
-        bert,
-        ja_bert,
         sid,
         noise_scale,
         length_scale,
@@ -147,7 +159,11 @@ class ModelWrapper(torch.nn.Module):
           lang_id: A 1-D array of dtype np.int64. Its shape is (token_numbers,)
           sid: an integer
         """
-        return self.model.infer(
+        bert = torch.zeros(x.shape[0], 1024, x.shape[1], dtype=torch.float32)
+        ja_bert = torch.zeros(x.shape[0], 768, x.shape[1], dtype=torch.float32)
+        lang_id = torch.zeros_like(x)
+        lang_id[:, 1::2] = self.lang_id
+        return self.model.model.infer(
             x=x,
             x_lengths=x_lengths,
             sid=sid,
@@ -169,7 +185,7 @@ def main():
 
     generate_tokens(model.hps["symbols"])
 
-    torch_model = ModelWrapper(model.model)
+    torch_model = ModelWrapper(model)
 
     opset_version = 13
     x = torch.randint(low=0, high=10, size=(60,), dtype=torch.int64)
@@ -177,19 +193,13 @@ def main():
     x_lengths = torch.tensor([x.size(0)], dtype=torch.int64)
     sid = torch.tensor([1], dtype=torch.int64)
     tones = torch.zeros_like(x)
-    lang_id = torch.ones_like(x)
+
     noise_scale = torch.tensor([1.0], dtype=torch.float32)
     length_scale = torch.tensor([1.0], dtype=torch.float32)
     noise_scale_w = torch.tensor([1.0], dtype=torch.float32)
 
-    bert = torch.zeros(1024, x.shape[0], dtype=torch.float32)
-    ja_bert = torch.zeros(768, x.shape[0], dtype=torch.float32)
-
     x = x.unsqueeze(0)
     tones = tones.unsqueeze(0)
-    lang_id = lang_id.unsqueeze(0)
-    bert = bert.unsqueeze(0)
-    ja_bert = ja_bert.unsqueeze(0)
 
     filename = "model.onnx"
 
@@ -199,9 +209,6 @@ def main():
             x,
             x_lengths,
             tones,
-            lang_id,
-            bert,
-            ja_bert,
             sid,
             noise_scale,
             length_scale,
@@ -213,9 +220,6 @@ def main():
             "x",
             "x_lengths",
             "tones",
-            "lang_id",
-            "bert",
-            "ja_bert",
             "sid",
             "noise_scale",
             "length_scale",
@@ -226,9 +230,6 @@ def main():
             "x": {0: "N", 1: "L"},
             "x_lengths": {0: "N"},
             "tones": {0: "N", 1: "L"},
-            "lang_id": {0: "N", 1: "L"},
-            "bert": {0: "N", 2: "L"},
-            "ja_bert": {0: "N", 2: "L"},
             "y": {0: "N", 1: "S", 2: "T"},
         },
     )

scripts/melo-tts/run.sh

@@ -28,6 +28,8 @@ echo "pwd: $PWD"
 
 ls -lh
 
+./show-info.py
+
 head lexicon.txt
 echo "---"
 tail lexicon.txt

scripts/melo-tts/show-info.py

@@ -0,0 +1,50 @@
+#!/usr/bin/env python3
+# Copyright      2024  Xiaomi Corp.        (authors: Fangjun Kuang)
+
+import onnxruntime
+
+
+def show(filename):
+    session_opts = onnxruntime.SessionOptions()
+    session_opts.log_severity_level = 3
+    sess = onnxruntime.InferenceSession(filename, session_opts)
+    for i in sess.get_inputs():
+        print(i)
+
+    print("-----")
+
+    for i in sess.get_outputs():
+        print(i)
+
+    meta = sess.get_modelmeta().custom_metadata_map
+    print("*****************************************")
+    print("meta\n", meta)
+
+
+def main():
+    print("=========model==========")
+    show("./model.onnx")
+
+
+if __name__ == "__main__":
+    main()
+
+"""
+=========model==========
+NodeArg(name='x', type='tensor(int64)', shape=['N', 'L'])
+NodeArg(name='x_lengths', type='tensor(int64)', shape=['N'])
+NodeArg(name='tones', type='tensor(int64)', shape=['N', 'L'])
+NodeArg(name='sid', type='tensor(int64)', shape=[1])
+NodeArg(name='noise_scale', type='tensor(float)', shape=[1])
+NodeArg(name='length_scale', type='tensor(float)', shape=[1])
+NodeArg(name='noise_scale_w', type='tensor(float)', shape=[1])
+-----
+NodeArg(name='y', type='tensor(float)', shape=['N', 'S', 'T'])
+*****************************************
+meta
+ {'description': 'MeloTTS is a high-quality multi-lingual text-to-speech library by MyShell.ai',
+ 'model_type': 'melo-vits', 'license': 'MIT license', 'sample_rate': '44100', 'add_blank': '1',
+ 'n_speakers': '1', 'bert_dim': '1024', 'language': 'Chinese + English',
+ 'ja_bert_dim': '768', 'speaker_id': '1', 'comment': 'melo', 'lang_id': '3',
+ 'tone_start': '0', 'url': 'https://github.com/myshell-ai/MeloTTS'}
+"""

scripts/melo-tts/test.py

@@ -30,6 +30,8 @@ class Lexicon:
                 tones = [int(t) for t in tones]
 
                 lexicon[word_or_phrase] = (phones, tones)
+        lexicon["呣"] = lexicon["母"]
+        lexicon["嗯"] = lexicon["恩"]
         self.lexicon = lexicon
 
         punctuation = ["!", "?", "…", ",", ".", "'", "-"]
@@ -98,20 +100,16 @@ class OnnxModel:
         self.lang_id = int(meta["lang_id"])
         self.sample_rate = int(meta["sample_rate"])
 
-    def __call__(self, x, tones, lang):
+    def __call__(self, x, tones):
         """
         Args:
           x: 1-D int64 torch tensor
           tones: 1-D int64 torch tensor
-          lang: 1-D int64 torch tensor
         """
         x = x.unsqueeze(0)
         tones = tones.unsqueeze(0)
-        lang = lang.unsqueeze(0)
 
-        print(x.shape, tones.shape, lang.shape)
+        print(x.shape, tones.shape)
-        bert = torch.zeros(1, self.bert_dim, x.shape[-1])
-        ja_bert = torch.zeros(1, self.ja_bert_dim, x.shape[-1])
         sid = torch.tensor([self.speaker_id], dtype=torch.int64)
         noise_scale = torch.tensor([0.6], dtype=torch.float32)
         length_scale = torch.tensor([1.0], dtype=torch.float32)
@@ -125,9 +123,6 @@ class OnnxModel:
                 "x": x.numpy(),
                 "x_lengths": x_lengths.numpy(),
                 "tones": tones.numpy(),
-                "lang_id": lang.numpy(),
-                "bert": bert.numpy(),
-                "ja_bert": ja_bert.numpy(),
                 "sid": sid.numpy(),
                 "noise_scale": noise_scale.numpy(),
                 "noise_scale_w": noise_scale_w.numpy(),
@@ -140,34 +135,46 @@ 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。多音字测试， 银行，行不行？长沙长大"
+    text = "永远相信，美好的事情即将发生。"
     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")
-    langs = [model.lang_id] * len(phones)
 
     if model.add_blank:
         new_phones = [0] * (2 * len(phones) + 1)
         new_tones = [0] * (2 * len(tones) + 1)
-        new_langs = [0] * (2 * len(langs) + 1)
 
         new_phones[1::2] = phones
         new_tones[1::2] = tones
-        new_langs[1::2] = langs
 
         phones = new_phones
         tones = new_tones
-        langs = new_langs
 
     phones = torch.tensor(phones, dtype=torch.int64)
     tones = torch.tensor(tones, dtype=torch.int64)
-    langs = torch.tensor(langs, dtype=torch.int64)
 
-    print(phones.shape, tones.shape, langs.shape)
+    print(phones.shape, tones.shape)
 
-    y = model(x=phones, tones=tones, lang=langs)
+    y = model(x=phones, tones=tones)
     sf.write("./test.wav", y, model.sample_rate)