add WeTextProcessing

README.md

@@ -22,6 +22,8 @@ git submodule update --init --recursive
 ``` sh
 conda create -n cosyvoice python=3.8
 conda activate cosyvoice
+# pynini is required by WeTextProcessing, use conda to install it as it can be executed on all platform.
+conda install -y -c conda-forge pynini==2.1.5
 pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com
 
 # If you encounter sox compatibility issues
@@ -55,9 +57,9 @@ git clone https://www.modelscope.cn/iic/CosyVoice-300M-Instruct.git pretrained_m
 git clone https://www.modelscope.cn/iic/CosyVoice-ttsfrd.git pretrained_models/CosyVoice-ttsfrd
 ```
 
-Optionaly, you can unzip `ttsfrd` resouce and install `ttsfrd` package.
+Optionaly, you can unzip `ttsfrd` resouce and install `ttsfrd` package for better text normalization performance.
 
-Notice that this step is not necessary. If you do not install `ttsfrd` package, you need to normalize input text manually.
+Notice that this step is not necessary. If you do not install `ttsfrd` package, we will use WeTextProcessing by default.
 
 ``` sh
 cd pretrained_models/CosyVoice-ttsfrd/

cosyvoice/cli/frontend.py

@@ -21,13 +21,13 @@ import torchaudio.compliance.kaldi as kaldi
 import torchaudio
 import os
 import inflect
-from tn.chinese.normalizer import Normalizer as ZhNormalizer
-from tn.english.normalizer import Normalizer as EnNormalizer
 try:
     import ttsfrd
     use_ttsfrd = True
-except:
-    print("failed to import ttsfrd, please normalize input text manually")
+except ImportError:
+    print("failed to import ttsfrd, use WeTextProcessing instead")
+    from tn.chinese.normalizer import Normalizer as ZhNormalizer
+    from tn.english.normalizer import Normalizer as EnNormalizer
     use_ttsfrd = False
 from cosyvoice.utils.frontend_utils import contains_chinese, replace_blank, replace_corner_mark, remove_bracket, spell_out_number, split_paragraph
 
@@ -64,8 +64,8 @@ class CosyVoiceFrontEnd:
             self.frd.enable_pinyin_mix(True)
             self.frd.set_breakmodel_index(1)
         else:
-            self.zh_tn_model = ZhNormalizer(remove_erhua=False,full_to_half=False)
-            self.en_tn_model = EnNormalizer()	
+            self.zh_tn_model = ZhNormalizer(remove_erhua=False, full_to_half=False)
+            self.en_tn_model = EnNormalizer()
 
     def _extract_text_token(self, text):
         text_token = self.tokenizer.encode(text, allowed_special=self.allowed_special)
@@ -103,7 +103,7 @@ class CosyVoiceFrontEnd:
             if self.use_ttsfrd:
                 text = self.frd.get_frd_extra_info(text, 'input')
             else:
-                text = self.zh_tn_model.normalize(text)	
+                text = self.zh_tn_model.normalize(text)
             text = text.replace("\n", "")
             text = replace_blank(text)
             text = replace_corner_mark(text)

cosyvoice/dataset/processor.py

@@ -22,7 +22,6 @@ from torch.nn.utils.rnn import pad_sequence
 import torch.nn.functional as F
 
 torchaudio.set_audio_backend('soundfile')
-torchaudio.utils.sox_utils.set_buffer_size(16500)
 
 AUDIO_FORMAT_SETS = set(['flac', 'mp3', 'm4a', 'ogg', 'opus', 'wav', 'wma'])
 

requirements.txt

@@ -26,4 +26,4 @@ tensorboard==2.14.0
 torch==2.0.1
 torchaudio==2.0.2
 wget==3.2
-WeTextProcessing
+WeTextProcessing==1.0.3