热词模型微调过程中遇到的问题

[David19970306]

OS: linux
Python Version：3.7.16
Package Version：pytorch 1.13.1、funasr 0.7.0
Model：speech_paraformer-large-contextual_asr_nat-zh-cn-16k-common-vocab8404
Command：python finetune.py
Details：使用本地数据微调，微调过程中无法进行提督更新

[Lizerui9926]

Please check if the dataset is correct and if it is too long. The audio exceeding 15 seconds will be filtered by default during training.

[David19970306]

the lengths of my audios are less than 15s，and i had checked the filter wav text fun，it does not filter any audio Sent from my iPhone

…

------------------ 原始邮件 ------------------ From: Lizerui9926 ***@***.***> 发送时间: 07/20/2023, 20:09 To: alibaba-damo-academy/FunASR ***@***.***> 抄送: 吴伟鹏 ***@***.***>, Author ***@***.***> Subject: Re: [alibaba-damo-academy/FunASR] 热词模型微调过程中遇到的问题 (Issue #756) Please check if the dataset is correct and if it is too long. The audio exceeding 15 seconds will be filtered by default during training. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***>

[yata0]

请问热词版微调的数据格式应该是什么样子的

[Lizerui9926]

请问热词版微调的数据格式应该是什么样子的

The training data format is the same as base paraformer, you can refer to https://alibaba-damo-academy.github.io/FunASR/en/egs_modelscope/asr/TEMPLATE/README.html#finetune-with-your-data