haha010508
haha010508
Thanks for your reply, but the wer_score=1.0, is mean totally wrong word, i do not think so, for example, ground_truths = 宋朝末年年间定居粉岭围, hypothesis = "宋朝末年年间定居分定为" only last 3 words is...
yes, CER is correct, i just focused on WER yesterday, because CER is confuse for chinese words.
Now, I'm sure mysql-connector-Python is causing a memory leak. If I keep running recognize_files (there are multiple audio files), the Memory of Python processes keeps getting higher and higher until...
This is my fault, i open plot and forget to close, so everytime 10M memery is lost.
I have the same problem, so i want to know, do you find the method for this?
> 网路问题，建议使用畅通稳定的网络下载，vox2数据集非常大。或者搜索一下第三方链接。 不是网络问题，好像就是不提供了，你可以试试，vox1我是可以下载的
我在广告中加了几条自己的数据，结果完全没学到，预测的还是广告的东西，完全不可用
> https://github.com/liucongg/ChatGLM-Finetuning 这个号称调出来了。我们没有验证过，你可以试试
i use the pretrained model, and same database(Dataset-mini), for evalue step, but i got very poor result, i want to know: why? this is my code ` CUDA_VISIBLE_DEVICES=1 python run.py...
i found this method will load two model, one is s3prl upstream model, one is wavlm_large finetune model, total size is 2.4G, how to reduce the size?