Merge pull request #83 from Artrajz/dev

Change docker base image

Dockerfile

@@ -1,12 +1,14 @@
-FROM python:3.10.11-slim-bullseye
+FROM artrajz/pytorch:1.13.1-cpu-py3.10.11-ubuntu22.04
 
 RUN mkdir -p /app
 WORKDIR /app
 
 ENV DEBIAN_FRONTEND=noninteractive
 
+
 RUN apt-get update && \
-    apt-get install -yq build-essential espeak-ng cmake wget && \
+    apt-get install -yq build-essential espeak-ng cmake wget ca-certificates tzdata&& \
+    update-ca-certificates && \
     apt-get clean && \
     apt-get purge -y --auto-remove -o APT::AutoRemove::RecommendsImportant=false && \
     rm -rf /var/lib/apt/lists/* 
@@ -16,20 +18,16 @@ RUN wget https://github.com/jemalloc/jemalloc/releases/download/5.3.0/jemalloc-5
     tar -xvf jemalloc-5.3.0.tar.bz2 && \
     cd jemalloc-5.3.0 && \
     ./configure && \
-    make && \
+    make -j$(nproc) && \
     make install && \
     cd .. && \
     rm -rf jemalloc-5.3.0* && \
-    ldconfig 
+    ldconfig
 
 ENV LD_PRELOAD=/usr/local/lib/libjemalloc.so
 
-RUN pip install torch --index-url https://download.pytorch.org/whl/cpu --no-cache-dir
-
 COPY requirements.txt /app/
-RUN pip install --upgrade pip && \
-    pip install pyopenjtalk==0.3.2 -i https://pypi.artrajz.cn/simple --no-cache-dir && \
-    pip install gunicorn --no-cache-dir && \
+RUN pip install gunicorn --no-cache-dir && \
     pip install -r requirements.txt --no-cache-dir&& \
     rm -rf /root/.cache/pip/*
 

Dockerfile_GPU

@@ -1,12 +1,12 @@
-FROM cnstark/pytorch:2.0.1-py3.10.11-cuda11.8.0-ubuntu22.04
+FROM artrajz/pytorch:1.13.1-cu117-py3.10.11-ubuntu22.04
 
 RUN mkdir -p /app
 WORKDIR /app
 
 ENV DEBIAN_FRONTEND=noninteractive
 
 RUN apt-get update && \
-    apt-get install -yq build-essential espeak-ng cmake wget ca-certificates && \
+    apt-get install -yq build-essential espeak-ng cmake wget ca-certificates tzdata&& \
     update-ca-certificates && \
     apt-get clean && \
     apt-get purge -y --auto-remove -o APT::AutoRemove::RecommendsImportant=false && \
@@ -18,18 +18,16 @@ RUN wget https://github.com/jemalloc/jemalloc/releases/download/5.3.0/jemalloc-5
     tar -xvf jemalloc-5.3.0.tar.bz2 && \
     cd jemalloc-5.3.0 && \
     ./configure && \
-    make && \
+    make -j$(nproc) && \
     make install && \
     cd .. && \
     rm -rf jemalloc-5.3.0* && \
-    ldconfig 
+    ldconfig
 
 ENV LD_PRELOAD=/usr/local/lib/libjemalloc.so
 
 COPY requirements.txt /app/
-RUN pip install --upgrade pip && \
-    pip install pyopenjtalk==0.3.2 fasttext -i https://pypi.artrajz.cn/simple --no-cache-dir && \
-    pip install gunicorn --no-cache-dir && \
+RUN pip install gunicorn --no-cache-dir && \
     pip install -r requirements.txt --no-cache-dir&& \
     rm -rf /root/.cache/pip/*
 

docker-compose-gpu.yaml

@@ -8,7 +8,7 @@ services:
       - 23456:23456
     environment:
       LANG: 'C.UTF-8'
-      #TZ: Asia/Shanghai #timezone
+      TZ: Asia/Shanghai #timezone
     volumes:
       - ./Model:/app/Model # 挂载模型文件夹
       - ./config.py:/app/config.py # 挂载配置文件

vits/text/japanese.py

@@ -1,7 +1,23 @@
+import os
 import re
 from unidecode import unidecode
 import pyopenjtalk
 
+from config import ABS_PATH
+from utils.download import download_and_verify
+
+URLS = [
+    "https://github.com/r9y9/open_jtalk/releases/download/v1.11.1/open_jtalk_dic_utf_8-1.11.tar.gz",
+    "https://ghproxy.com/https://github.com/r9y9/open_jtalk/releases/download/v1.11.1/open_jtalk_dic_utf_8-1.11.tar.gz",
+]
+install_path = os.path.dirname(pyopenjtalk.__file__)
+dict_path = os.path.join(install_path, "open_jtalk_dic_utf_8-1.11", "char.bin")
+TARGET_PATH = os.path.join(ABS_PATH, "open_jtalk_dic_utf_8-1.11.tar.gz")
+EXTRACT_DESTINATION = os.path.join(install_path, "")
+EXPECTED_MD5 = None
+
+if not os.path.exists(dict_path):
+    success, message = download_and_verify(URLS, TARGET_PATH, EXPECTED_MD5, EXTRACT_DESTINATION)
 
 # Regular expression matching Japanese without punctuation marks:
 _japanese_characters = re.compile(
@@ -127,7 +143,7 @@ def get_real_hatsuon(text):
 def japanese_to_ipa(text):
     text = japanese_to_romaji_with_accent(text).replace('...', '…')
     text = re.sub(
-        r'([aiueo])\1+', lambda x: x.group(0)[0]+'ː'*(len(x.group(0))-1), text)
+        r'([aiueo])\1+', lambda x: x.group(0)[0] + 'ː' * (len(x.group(0)) - 1), text)
     text = get_real_sokuon(text)
     text = get_real_hatsuon(text)
     for regex, replacement in _romaji_to_ipa:
@@ -148,6 +164,6 @@ def japanese_to_ipa3(text):
     text = japanese_to_ipa2(text).replace('n^', 'ȵ').replace(
         'ʃ', 'ɕ').replace('*', '\u0325').replace('#', '\u031a')
     text = re.sub(
-        r'([aiɯeo])\1+', lambda x: x.group(0)[0]+'ː'*(len(x.group(0))-1), text)
+        r'([aiɯeo])\1+', lambda x: x.group(0)[0] + 'ː' * (len(x.group(0)) - 1), text)
     text = re.sub(r'((?:^|\s)(?:ts|tɕ|[kpt]))', r'\1ʰ', text)
     return text