update dockerfile

cosyvoice/dataset/processor.py

@@ -106,7 +106,7 @@ def filter(data,
         yield sample
 
 
-def resample(data, resample_rate=22050, mode='train'):
+def resample(data, resample_rate=22050, min_sample_rate=16000, mode='train'):
     """ Resample data.
         Inplace operation.
 
@@ -123,7 +123,7 @@ def resample(data, resample_rate=22050, mode='train'):
         sample_rate = sample['sample_rate']
         waveform = sample['speech']
         if sample_rate != resample_rate:
-            if sample_rate < resample_rate:
+            if sample_rate < min_sample_rate:
                 continue
             sample['sample_rate'] = resample_rate
             sample['speech'] = torchaudio.transforms.Resample(

runtime/python/Dockerfile

@@ -1,15 +1,14 @@
-FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04
+FROM pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime
 ENV DEBIAN_FRONTEND=noninteractive
 
 WORKDIR /opt/CosyVoice
 
 RUN sed -i s@/archive.ubuntu.com/@/mirrors.aliyun.com/@g /etc/apt/sources.list
 RUN apt-get update -y
-RUN apt-get -y install python3-dev cmake python3-pip git unzip
+RUN apt-get -y install git unzip git-lfs
+RUN git lfs install
 RUN git clone --recursive https://github.com/FunAudioLLM/CosyVoice.git
-RUN cd CosyVoice && pip3 install --default-timeout=3600 -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com
-RUN apt install git-lfs && git lfs install
-RUN cd CosyVoice && git clone https://www.modelscope.cn/iic/CosyVoice-ttsfrd.git pretrained_models/CosyVoice-ttsfrd
-RUN cd CosyVoice/pretrained_models/CosyVoice-ttsfrd && unzip resource.zip -d . && pip3 install ttsfrd-0.3.6-cp38-cp38-linux_x86_64.whl
+# here we use python==3.10 because we cannot find an image which have both python3.8 and torch2.0.1-cu118 installed
+RUN cd CosyVoice && pip3 install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com
 RUN cd CosyVoice/runtime/python && python3 -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. cosyvoice.proto
-CMD ["/bin/bash", "-c", "cd /opt/CosyVoice/CosyVoice/runtime/python && . ./path/sh && python3 server.py --port 50000 --max_conc 4 --model_dir iic/CosyVoice-300M && sleep infinity"]
+CMD ["/bin/bash", "-c", "cd /opt/CosyVoice/CosyVoice/runtime/python && python3 server.py --port 50000 --max_conc 4 --model_dir iic/CosyVoice-300M && sleep infinity"]

webui.py

@@ -59,10 +59,10 @@ def postprocess(speech, top_db=60, hop_length=220, win_length=440):
     return speech
 
 inference_mode_list = ['预训练音色', '3s极速复刻', '跨语种复刻', '自然语言控制']
-instruct_dict = {'预训练音色': '1. 选择预训练音色\n2.点击生成音频按钮',
-                 '3s极速复刻': '1. 选择prompt音频文件，或录入prompt音频，若同时提供，优先选择prompt音频文件\n2. 输入prompt文本\n3.点击生成音频按钮',
-                 '跨语种复刻': '1. 选择prompt音频文件，或录入prompt音频，若同时提供，优先选择prompt音频文件\n2.点击生成音频按钮',
-                 '自然语言控制': '1. 输入instruct文本\n2.点击生成音频按钮'}
+instruct_dict = {'预训练音色': '1. 选择预训练音色\n2. 点击生成音频按钮',
+                 '3s极速复刻': '1. 选择prompt音频文件，或录入prompt音频，注意不超过30s，若同时提供，优先选择prompt音频文件\n2. 输入prompt文本\n3. 点击生成音频按钮',
+                 '跨语种复刻': '1. 选择prompt音频文件，或录入prompt音频，注意不超过30s，若同时提供，优先选择prompt音频文件\n2. 点击生成音频按钮',
+                 '自然语言控制': '1. 选择预训练音色\n2. 输入instruct文本\n3. 点击生成音频按钮'}
 def change_instruction(mode_checkbox_group):
     return instruct_dict[mode_checkbox_group]