关于FastSpeech2 with CSMSC训练 #2349
-
跑到这一步时 |
Beta Was this translation helpful? Give feedback.
Answered by
yt605155624
Sep 6, 2022
Replies: 1 comment
-
我在 2022.09.05 拉取的 paddle develop 和 paddle2onnx 1.0.0 下,用预训练模型执行如下命令,无法复现你的问题 wget https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_cnndecoder_csmsc_ckpt_1.0.0.zip
unzip fastspeech2_cnndecoder_csmsc_ckpt_1.0.0.zip
mkdir -p exp/cnndecoder/checkpoints
cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/snapshot_iter_153000.pdz exp/cnndecoder/checkpoints
mkdir -p dump/train
cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/speech_stats.npy dump/train
cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/phone_id_map.txt dump
# 修改 run_cnndecoder.sh 里的 ckpt_name
wget https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_ckpt_0.4.zip
unzip pwg_baker_ckpt_0.4.zip
./run_cnndecoder.sh --stage 3 --stop-stage 10 你在 86 行报错是 am warm up 时就有问题,可能是你自己训练的和我们的参数不匹配, 此处的输入数值是根据我们的预训练模型定义的,如 phone_id_map.txt 有 268 行所以此处选择 266,你需要按照自己的模型定义这里的数值, 可以用 https://netron.app/ 可视化你导出的 onnx 模型确定参数类型和范围 |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
yt605155624
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
我在 2022.09.05 拉取的 paddle develop 和 paddle2onnx 1.0.0 下,用预训练模型执行如下命令,无法复现你的问题
wget https://paddlespeech.bj.bcebos.com/Parakeet/released_models/fastspeech2/fastspeech2_cnndecoder_csmsc_ckpt_1.0.0.zip unzip fastspeech2_cnndecoder_csmsc_ckpt_1.0.0.zip mkdir -p exp/cnndecoder/checkpoints cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/snapshot_iter_153000.pdz exp/cnndecoder/checkpoints mkdir -p dump/train cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/speech_stats.npy dump/train cp fastspeech2_cnndecoder_csmsc_ckpt_1.0.0/phone_id_map.txt dump # 修改 run_cnndecoder.sh 里的 ckpt_name wget https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_ckpt_0.4.zip unzip pwg_baker_ckpt_0.4.zip ./r…