FunASR
FunASR copied to clipboard
关于使用越南语进行微调,不收敛的问题
trafficstars
Notice: In order to resolve issues more efficiently, please raise issue following the template.
(注意:为了更加高效率解决您遇到的问题,请按照模板提问,补充细节)
我使用了UniASR的实时越南语模型,https://modelscope.cn/models/iic/speech_UniASR_asr_2pass-vi-16k-common-vocab1001-pytorch-online/summary
因为UniASR下面目前还没更新微调和训练的脚本,我使用的是https://github.com/alibaba-damo-academy/FunASR/tree/main/examples/industrial_data_pretraining/paraformer,的微调脚本进行训练的,训练参数保持和脚本中的一致,改了一下预训练的模型为UniASR中的越南语模型,训练了100epoch之后,准确率一直保持在80%左右
❓ Questions and Help
我是否需要重新去生成tokens和cmvn文件进行重新训练?
Before asking:
- search the issues.
- search the docs.
What is your question?
Code
What have you tried?
What's your environment?
- OS (e.g., Linux): Linux
- FunASR Version (e.g., 1.0.0): 1.0.20
- ModelScope Version (e.g., 1.11.0):1.13.3
- PyTorch Version (e.g., 2.0.0):2.2.2
- How you installed funasr (
pip, source):pip - Python version:
- GPU (e.g., V100M32):RTX3090
- CUDA/cuDNN version (e.g., cuda11.7):11.8