search

FunAudioLLM / CosyVoice

Multi-lingual large voice generation model, providing inference, training and deployment full-stack ability.
https://funaudiollm.github.io/
Apache License 2.0
5.74k stars 610 forks source link

准备数据parquet的时间很长 #214

Open zuotianlun opened 2 months ago

zuotianlun commented 2 months ago

我想微调cosyvoice，但我发现当我给定num_utts_per_parquet为10000时，生成的parquet_0XXX.tar文件非常大，约17G，而且生成时间很长，生成一个tar文件大概要几十分钟，请问这是正常的吗

aluminumbox commented 2 months ago

check utt_embedding in parquet_0XXX.tar. in latest code, utt_embedding is a list with len 192. in libritts, each parquet file is approximately 300M, I think your parquet file is a bit too large