masterwang22327 comments

Results 15 comments of


                                            masterwang22327

Can I replace the espeak-ng phonemizer with neural g2p

waitting for your work.i noticed the same issu,some missed phoneme can be fix by espeak-ng.https://github.com/rhasspy/piper/issues/344#issuecomment-1890844487

why trained model size is about 500M,but pretrained model size is 200M?

> it's with the optimizers yeah,i found this thing,and reduce the no-needed paramater at end.thank u.

Problem with training

> modify the train.sh to below and remember let your train dataset 44.1k pcm ` CONFIG=$1 GPUS=$2 MODEL_NAME=$(basename "$(dirname $CONFIG)") > > PORT=10902 > > torchrun --nproc_per_node=$GPUS --master_port=$PORT train.py --c...

Problem with training

I have the same issue.But i am not fix it,the model is also trained good.But the trained model size is about 500M,the official pretrained model size is 200M.where the problem...

embedding model前向推理时，不同batchsize，相同文本，输出embedding小数点后几位不同

我也遇到了一样的问题批量请求的结果和单条请求的向量小数位最后几位不一致导致量化的时候也会有点问题进而影响查询精度