Skip to content

Cosyvoice3:方言微调的效果反而变差了 #1839

@zyy-fc

Description

@zyy-fc

基于Cosyvoice3的llm模型做粤语的微调,目的是期望进一步降低粤语的wer

训练数据是从WenetSpeech-Yue中挑选出高质量的数据,分别有10小时和200小时两种,但是微调的结果都变差了,幻觉问题很严重,想问问大家是怎么解决的呢?

instruct采用的是“请用广东话表达”

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions