Merge pull request babysor#13 from babysor/chineseinputsupport

Toolbox add Chinese character input support

README-CN.md

@@ -45,6 +45,8 @@
 然后您可以尝试使用工具箱：
 `python demo_toolbox.py -d <datasets_root>`
 
+> Good news🤩: 可直接使用中文
+
 ## TODO
 - [X] 允许直接使用中文
 - [X] 添加演示视频

README.md

@@ -51,6 +51,8 @@ You can then try the toolbox:
 or  
 `python demo_toolbox.py`  
 
+> Good news🤩: Chinese Characters are supported
+
 ## TODO
 - [x] Add demo video
 - [X] Add support for more dataset

synthesizer/inference.py

@@ -9,7 +9,7 @@
 from typing import Union, List
 import numpy as np
 import librosa
-
+from pypinyin import lazy_pinyin, Style
 
 class Synthesizer:
     sample_rate = hparams.sample_rate
@@ -91,6 +91,10 @@ def synthesize_spectrograms(self, texts: List[str],
             simple_table([("Tacotron", str(tts_k) + "k"),
                         ("r", self._model.r)])
 
+        #convert chinese char to pinyin
+        list_of_pinyin = lazy_pinyin(texts, style=Style.TONE3)
+        texts = [" ".join([v for v in list_of_pinyin if v.strip()])]
+
         # Preprocess text inputs
         inputs = [text_to_sequence(text.strip(), hparams.tts_cleaner_names) for text in texts]
         if not isinstance(embeddings, list):

toolbox/ui.py

@@ -36,17 +36,8 @@
 ], dtype=np.float) / 255 
 
 default_text = \
-    "Welcome to the toolbox! To begin, load an utterance from your datasets or record one " \
-    "yourself.\nOnce its embedding has been created, you can synthesize any text written here.\n" \
-    "The synthesizer expects to generate " \
-    "outputs that are somewhere between 5 and 12 seconds.\nTo mark breaks, write a new line. " \
-    "Each line will be treated separately.\nThen, they are joined together to make the final " \
-    "spectrogram. Use the vocoder to generate audio.\nThe vocoder generates almost in constant " \
-    "time, so it will be more time efficient for longer inputs like this one.\nOn the left you " \
-    "have the embedding projections. Load or record more utterances to see them.\nIf you have " \
-    "at least 2 or 3 utterances from a same speaker, a cluster should form.\nSynthesized " \
-    "utterances are of the same color as the speaker whose voice was used, but they're " \
-    "represented with a cross."
+    "欢迎使用工具箱, 现已支持中文输入！"
+
 
 
 class UI(QDialog):