Xingchen Song(宋星辰) comments

Results 173 comments of


                                            Xingchen Song(宋星辰)

Getting decoding output <unk> <unk> <unk> <unk>

u may refer to - https://github.com/wenet-e2e/wenet/issues/825 - https://github.com/wenet-e2e/wenet/issues/519 - https://github.com/wenet-e2e/wenet/issues/1673

Getting decoding output <unk> <unk> <unk> <unk>

and - https://github.com/wenet-e2e/wenet/issues/1545

Trained checkpoint produces empty result during runtime decoding

maye some audios have 2 channels, u can refer to https://github.com/wenet-e2e/wenet/issues/519

continuous_decoding mode seems to miss alternate segments.

> memory leak has been fixed by jemalloc, see https://github.com/wenet-e2e/wenet/issues/352

continuous_decoding mode seems to miss alternate segments.

> > I think this is due to vad info is depends on the blank prob. > > In ResetContinuousDecoding we should keep the state of some recent frames, not...

Missing word problem

try to pad zero at the front of wav

Random output with WFST decoding

please follow https://github.com/wenet-e2e/wenet/issues/1673

多 GPU 训练在第二个 epoch 卡在 loss.backward()

torch 的 join() 以及 cv的static 可以解决这个问题，see https://github.com/wenet-e2e/wenet/issues/842 and https://github.com/wenet-e2e/wenet/pull/903

uing shard data type for librispeech, I got errors " WARNING error to parse"

u can pre-process xx.flac to xx.wav before training recipe

[runtime] add jemalloc

close this PR and leave it as a reference