fine-tuned with japanese

.gitignore

@@ -5,6 +5,8 @@
 /training/*
 /config/*
 /dataset/*
+/results/*
+/trained/*
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]

README.md

@@ -31,4 +31,5 @@
 -   Trim Segments: 使用TorchAudio的VAD来修剪静音，减小音频文件的实际大小，从而在训练时节省更多的处理时间和VRAM。
 -   Slice Start Offset: 在分割音频文件时偏移起始时间戳。
 -   Slice End Offset: 在分割音频文件时偏移结束时间戳。
--   
+-   Transcribe and Process: 开始转录，同时在必要时进行切片，并分割为训练集和验证集
+-   

dlas/data/audio/test.py

@@ -0,0 +1,3 @@
+import sys
+sys.path.append('/home/sjx/Common/vits/ai-voice-cloning/dlas')
+from pyfastmp3decoder.mp3decoder import load_mp3

req.txt

@@ -0,0 +1,4 @@
+pyyaml
+tb-nightly
+future
+munch

src/test.py

@@ -1,4 +1,2 @@
-import sys
-
-sys.path.append('/home/sjx/Common/vits/ai-voice-cloning')
-from dlas import train as tr
+import torch
+print(torch.distributed.get_world_size())

src/train.py

@@ -58,6 +58,8 @@ def train(config_path, launcher='none'):
         pass
 
     import torch
+    import sys
+    sys.path.append('/home/sjx/Common/vits/ai-voice-cloning')
     from dlas import train as tr
     from dlas.utils import util, options as option
 

train.sh

@@ -1,4 +1,5 @@
 #!/bin/bash
-conda activate tortoise
+# conda activate tortoise
+source /home/sjx/anaconda3/bin/activate /home/sjx/anaconda3/envs/tortoise 
 python3 ./src/train.py --yaml "$1"
-deactivate
+# deactivate