Merge branch 'main' into 添加推理页面的引导音频转写功能

Docker/download.py

@@ -0,0 +1,7 @@
+# Download moda ASR related models
+from modelscope import snapshot_download
+model_dir = snapshot_download('damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch')
+model_dir = snapshot_download('damo/speech_fsmn_vad_zh-cn-16k-common-pytorch')
+model_dir = snapshot_download('damo/punc_ct-transformer_zh-cn-common-vocab272727-pytorch')
+
+

Dockerfile

@@ -2,7 +2,7 @@
 FROM cnstark/pytorch:2.0.1-py3.9.17-cuda11.8.0-ubuntu20.04
 
 LABEL maintainer="breakstring@hotmail.com"
-LABEL version="dev-20240123.03"
+LABEL version="dev-20240127"
 LABEL description="Docker image for GPT-SoVITS"
 
 
@@ -18,27 +18,19 @@ RUN apt-get update && \
 WORKDIR /workspace
 COPY . /workspace
 
+# install python packages
+RUN pip install -r requirements.txt
+
 # Download models
 RUN chmod +x /workspace/Docker/download.sh && /workspace/Docker/download.sh
 
-# 本应该从 requirements.txt 里面安装package，但是由于funasr和modelscope的问题，暂时先在后面手工安装依赖包吧
-RUN pip install --no-cache-dir torch numpy scipy tensorboard librosa==0.9.2 numba==0.56.4 pytorch-lightning gradio==3.14.0 ffmpeg-python onnxruntime tqdm cn2an pypinyin pyopenjtalk g2p_en chardet transformers jieba psutil PyYAML
-# 这里强制指定了modelscope和funasr的版本，后面damo_asr的模型让它们自己下载
-RUN pip install --no-cache-dir modelscope~=1.10.0 torchaudio sentencepiece funasr~=0.8.7
+# Download moda ASR related
+RUN python /workspace/Docker/download.py
 
-# 先屏蔽掉，让容器里自己下载
-# Clone damo_asr
-#WORKDIR /workspace/tools/damo_asr/models
-#RUN git clone --depth 1 https://www.modelscope.cn/iic/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch.git speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch && \
-#    (cd speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch && git lfs pull)
-#RUN git clone --depth 1 https://www.modelscope.cn/iic/speech_fsmn_vad_zh-cn-16k-common-pytorch.git speech_fsmn_vad_zh-cn-16k-common-pytorch && \
-#    (cd speech_fsmn_vad_zh-cn-16k-common-pytorch && git lfs pull)
-#RUN git clone --depth 1 https://www.modelscope.cn/iic/punc_ct-transformer_zh-cn-common-vocab272727-pytorch.git punc_ct-transformer_zh-cn-common-vocab272727-pytorch && \
-#    (cd punc_ct-transformer_zh-cn-common-vocab272727-pytorch && git lfs pull)
+# Download nltk realted
+RUN python -m nltk.downloader averaged_perceptron_tagger
+RUN python -m nltk.downloader cmudict
 
-#RUN parallel --will-cite -a /workspace/Docker/damo.sha256 "echo -n {} | sha256sum -c"
-
-#WORKDIR /workspace
 
 EXPOSE 9870
 EXPOSE 9871

GPT_SoVITS/AR/data/data_module.py

@@ -41,7 +41,7 @@ class Text2SemanticDataModule(LightningDataModule):
         #     pad_val=self.config['data']['pad_val'])
 
     def train_dataloader(self):
-        batch_size = self.config["train"]["batch_size"]
+        batch_size = max(min(self.config["train"]["batch_size"],len(self._train_dataset)//4),1)#防止不保存
         sampler = DistributedBucketSampler(self._train_dataset, batch_size=batch_size)
         return DataLoader(
             self._train_dataset,

GPT_SoVITS/inference_webui.py

@@ -283,17 +283,26 @@ def nonen_get_bert_inf(text, language):
 
     return bert
 
-#i18n("不切"),i18n("凑五句一切"),i18n("凑50字一切"),i18n("按中文句号。切"),i18n("按英文句号.切")
+splits = {"，","。","？","！",",",".","?","!","~",":","：","—","…",}
+def get_first(text):
+    pattern = "[" + "".join(re.escape(sep) for sep in splits) + "]"
+    text = re.split(pattern, text)[0].strip()
+    return text
+
 def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,how_to_cut=i18n("不切")):
     t0 = ttime()
     prompt_text = prompt_text.strip("\n")
-    prompt_language, text = prompt_language, text.strip("\n")
+    if(prompt_text[-1]not in splits):prompt_text+="。"if prompt_text!="en"else "."
+    text = text.strip("\n")
+    if(len(get_first(text))<4):text+="。"if text!="en"else "."
     zero_wav = np.zeros(
         int(hps.data.sampling_rate * 0.3),
         dtype=np.float16 if is_half == True else np.float32,
     )
     with torch.no_grad():
         wav16k, sr = librosa.load(ref_wav_path, sr=16000)
+        if(wav16k.shape[0]>160000 or wav16k.shape[0]<48000):
+            raise OSError(i18n("参考音频在3~10秒范围外，请更换！"))
         wav16k = torch.from_numpy(wav16k)
         zero_wav_torch = torch.from_numpy(zero_wav)
         if is_half == True:
@@ -318,7 +327,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         phones1, word2ph1, norm_text1 = clean_text_inf(prompt_text, prompt_language)
     else:
         phones1, word2ph1, norm_text1 = nonen_clean_text_inf(prompt_text, prompt_language)
-    if(how_to_cut==i18n("凑五句一切")):text=cut1(text)
+    if(how_to_cut==i18n("凑四句一切")):text=cut1(text)
     elif(how_to_cut==i18n("凑50字一切")):text=cut2(text)
     elif(how_to_cut==i18n("按中文句号。切")):text=cut3(text)
     elif(how_to_cut==i18n("按英文句号.切")):text=cut4(text)
@@ -390,24 +399,6 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         np.int16
     )
 
-
-splits = {
-    "，",
-    "。",
-    "？",
-    "！",
-    ",",
-    ".",
-    "?",
-    "!",
-    "~",
-    ":",
-    "：",
-    "—",
-    "…",
-}  # 不考虑省略号
-
-
 def split(todo_text):
     todo_text = todo_text.replace("……", "。").replace("——", "，")
     if todo_text[-1] not in splits:
@@ -430,7 +421,7 @@ def split(todo_text):
 def cut1(inp):
     inp = inp.strip("\n")
     inps = split(inp)
-    split_idx = list(range(0, len(inps), 5))
+    split_idx = list(range(0, len(inps), 4))
     split_idx[-1] = None
     if len(split_idx) > 1:
         opts = []
@@ -444,7 +435,6 @@ def cut1(inp):
 def cut2(inp):
     inp = inp.strip("\n")
     inps = split(inp)
-    # print(inps)
     if len(inps) < 2:
         return inp
     opts = []
@@ -517,6 +507,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
         with gr.Row():
             inp_ref = gr.Audio(label=i18n("请上传参考音频"), type="filepath")
             whisper_button = gr.Button(i18n("faster_whisper转写音频内容到文本"))
+            inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath")
             prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="")
             prompt_language = gr.Dropdown(
                 label=i18n("参考音频的语种"),choices=[i18n("中文"),i18n("英文"),i18n("日文")],value=i18n("中文")
@@ -530,7 +521,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             )
             how_to_cut = gr.Radio(
                 label=i18n("怎么切"),
-                choices=[i18n("不切"),i18n("凑五句一切"),i18n("凑50字一切"),i18n("按中文句号。切"),i18n("按英文句号.切"),],
+                choices=[i18n("不切"),i18n("凑四句一切"),i18n("凑50字一切"),i18n("按中文句号。切"),i18n("按英文句号.切"),],
                 value=i18n("凑50字一切"),
                 interactive=True,
             )
@@ -546,7 +537,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
         gr.Markdown(value=i18n("文本切分工具。太长的文本合成出来效果不一定好，所以太长建议先切。合成会根据文本的换行分开合成再拼起来。"))
         with gr.Row():
             text_inp = gr.Textbox(label=i18n("需要合成的切分前文本"),value="")
-            button1 = gr.Button(i18n("凑五句一切"), variant="primary")
+            button1 = gr.Button(i18n("凑四句一切"), variant="primary")
             button2 = gr.Button(i18n("凑50字一切"), variant="primary")
             button3 = gr.Button(i18n("按中文句号。切"), variant="primary")
             button4 = gr.Button(i18n("按英文句号.切"), variant="primary")

GPT_SoVITS/s1_train.py

@@ -44,9 +44,8 @@ class my_model_ckpt(ModelCheckpoint):
         self.config = config
 
     def on_train_epoch_end(self, trainer, pl_module):
-        if not self._should_skip_saving_checkpoint(
-            trainer
-        ) and self._should_save_on_train_epoch_end(trainer):
+        # if not self._should_skip_saving_checkpoint(trainer) and self._should_save_on_train_epoch_end(trainer):
+        if self._should_save_on_train_epoch_end(trainer):
             monitor_candidates = self._monitor_candidates(trainer)
             if (
                 self._every_n_epochs >= 1

README.md

@@ -123,6 +123,7 @@ For UVR5 (Vocals/Accompaniment Separation & Reverberation Removal, additionally)
 
 #### docker-compose.yaml configuration 
 
+0. Regarding image tags: Due to rapid updates in the codebase and the slow process of packaging and testing images, please check [Docker Hub](https://hub.docker.com/r/breakstring/gpt-sovits) for the currently packaged latest images and select as per your situation, or alternatively, build locally using a Dockerfile according to your own needs.
 1. Environment Variables：
   - is_half: Controls half-precision/double-precision. This is typically the cause if the content under the directories 4-cnhubert/5-wav32k is not generated correctly during the "SSL extracting" step. Adjust to True or False based on your actual situation.
 
@@ -140,7 +141,7 @@ docker compose -f "docker-compose.yaml" up -d
 
 As above, modify the corresponding parameters based on your actual situation, then run the following command:
 ```
-docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:dev-20240123.03
+docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:xxxxx
 ```
 
 

docker-compose.yaml

@@ -2,7 +2,7 @@ version: '3.8'
 
 services:
   gpt-sovits:
-    image: breakstring/gpt-sovits:dev-20240123.03
+    image: breakstring/gpt-sovits:xxxxx   # please change the image name and tag base your environment
     container_name: gpt-sovits-container
     environment:
       - is_half=False

docs/cn/README.md

@@ -110,7 +110,7 @@ brew install ffmpeg
 ### 在 Docker 中使用
 
 #### docker-compose.yaml 设置
-
+0. image的标签：由于代码库更新很快，镜像的打包和测试又很慢，所以请自行在 [Docker Hub](https://hub.docker.com/r/breakstring/gpt-sovits) 查看当前打包好的最新的镜像并根据自己的情况选用，或者在本地根据您自己的需求通过Dockerfile进行构建。
 1. 环境变量：
   - is_half: 半精度/双精度控制。在进行 "SSL extracting" 步骤时如果无法正确生成 4-cnhubert/5-wav32k 目录下的内容时，一般都是它引起的，可以根据实际情况来调整为True或者False。
 
@@ -129,7 +129,7 @@ docker compose -f "docker-compose.yaml" up -d
 
 同上，根据您自己的实际情况修改对应的参数，然后运行如下命令：
 ```
-docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:dev-20240123.03
+docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:xxxxx
 ```
 
 

docs/ja/README.md

@@ -108,6 +108,7 @@ brew install ffmpeg
 
 #### docker-compose.yamlの設定 
 
+0. イメージのタグについて：コードベースの更新が速く、イメージのパッケージングとテストが遅いため、[Docker Hub](https://hub.docker.com/r/breakstring/gpt-sovits) で現在パッケージされている最新のイメージをご覧になり、ご自身の状況に応じて選択するか、またはご自身のニーズに応じてDockerfileを使用してローカルで構築してください。
 1. 環境変数：
     - `is_half`：半精度／倍精度の制御。"SSL抽出"ステップ中に`4-cnhubert/5-wav32k`ディレクトリ内の内容が正しく生成されない場合、通常これが原因です。実際の状況に応じてTrueまたはFalseに調整してください。
 
@@ -124,7 +125,7 @@ docker compose -f "docker-compose.yaml" up -d
 
 上記と同様に、実際の状況に基づいて対応するパラメータを変更し、次のコマンドを実行します：
 ```markdown
-docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:dev-20240123.03
+docker run --rm -it --gpus=all --env=is_half=False --volume=G:\GPT-SoVITS-DockerTest\output:/workspace/output --volume=G:\GPT-SoVITS-DockerTest\logs:/workspace/logs --volume=G:\GPT-SoVITS-DockerTest\SoVITS_weights:/workspace/SoVITS_weights --workdir=/workspace -p 9870:9870 -p 9871:9871 -p 9872:9872 -p 9873:9873 -p 9874:9874 --shm-size="16G" -d breakstring/gpt-sovits:xxxxx
 ```
 
 

webui.py

@@ -16,7 +16,11 @@ if(os.path.exists(tmp)):
         if(name=="jieba.cache"):continue
         path="%s/%s"%(tmp,name)
         delete=os.remove if os.path.isfile(path) else shutil.rmtree
+        try:
             delete(path)
+        except Exception as e:
+            print(str(e))
+            pass
 import site
 site_packages_roots = []
 for path in site.getsitepackages():
@@ -26,7 +30,6 @@ if(site_packages_roots==[]):site_packages_roots=["%s/runtime/Lib/site-packages"
 #os.environ["OPENBLAS_NUM_THREADS"] = "4"
 os.environ["no_proxy"] = "localhost, 127.0.0.1, ::1"
 os.environ["all_proxy"] = ""
-
 for site_packages_root in site_packages_roots:
     if os.path.exists(site_packages_root):
         try:
@@ -38,7 +41,6 @@ for site_packages_root in site_packages_roots:
             break
         except PermissionError:
             pass
-
 from tools import my_utils
 import traceback
 import shutil
@@ -662,7 +664,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             with gr.Row():
                 if_label = gr.Checkbox(label=i18n("是否开启打标WebUI"),show_label=True)
                 path_list = gr.Textbox(
-                    label=i18n("打标数据标注文件路径"),
+                    label=i18n(".list标注文件的路径"),
                     value="D:\\RVC1006\\GPT-SoVITS\\raw\\xxx.list",
                     interactive=True,
                 )
@@ -688,7 +690,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
                         label=i18n("*训练集音频文件目录"),
                         # value=r"D:\RVC1006\GPT-SoVITS\raw\xxx",
                         interactive=True,
-                        placeholder=i18n("训练集音频文件目录-拼接-list文件里波形对应的文件名（不是全路径）。")
+                        placeholder=i18n("填切割后音频所在目录！读取的音频文件完整路径=该目录-拼接-list文件里波形对应的文件名（不是全路径）。")
                     )
                 gr.Markdown(value=i18n("1Aa-文本内容"))
                 with gr.Row():