felixfuyihui / aishell-4 Goto Github PK

hi yihui, thanks for awesome system implementation,
I can only get the speaker info of the train set, but no speaker info for the test set.
http://aishell-4.oss-cn-hangzhou.aliyuncs.com/spk_info.xlsx
can you tell where I can find the test spk_info?

thanks a lot
@felixfuyihui

hi yihui！Thank you for your dedication to the code.But I have a few questions. In AISHELL-4/data_preparation/generate_fe_trainingdata.py line29、30 etc no random seed set in these places，also the documents in path/to/wavlist/of/speaker1、path/to/wavlist/of/speaker2 etc need to be prepared by myself, which makes it impossible to completely reproduce the results in the paper. How can I solve this problem?Thanks again!

• train_M/TextGrid/20200622_M_R002S07C01.TextGrid
• train_M/TextGrid/20200710_M_R002S06C01.TextGrid

这两条数据中 Item[x].xmax (2104.492) 应当等于全局的 xmax (2187.436)，否则 textgrid (version==1.5) 读取该文件会出错

出错示例：

It looks like the amount of non-overlapped data is much smaller than the overall corpus. I am seeing less than 20 hours. Is this correct?

Thanks
Michael Picheny

I tried to format all required .scp as . It worked for the front-end part, but I encountered some problems when doing ASR and evaluation. I wonder what the exact format of each .scp file is. especially wav_nospk_nofe.scp, wav_spk_nofe.scp, wav_nospk_fe.scp and wav_spk_fe.scp for ASR.
1~2 lines of examples should be very helpful.

Hi：
there is something wrong with the pre-trained asr model, it can't be decompressed, can you fixed this problem ????

Hi Fu, I notice scripts in data preparation have to take several wavlists as input, like spk1_list, wav_list, noise_list.
How can I get or generate these lists? Are there any rules or scripts? Thx!

I have got this error when train the model

Traceback (most recent call last):
File "steps/run_realmask.py", line 402, in
main()
File "steps/run_realmask.py", line 277, in main
train(model, device, writer)
File "steps/run_realmask.py", line 88, in train
val_loss = validation(model, -1, lr, device)
File "steps/run_realmask.py", line 209, in validation
for idx, data in enumerate(dataloader):
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 435, in next
data = self._next_data()
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1085, in _next_data
return self._process_data(data)
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1111, in _process_data
data.reraise()
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/_utils.py", line 428, in reraise
raise self.exc_type(msg)
RuntimeError: Caught RuntimeError in DataLoader worker process 0.

Original Traceback (most recent call last):
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 198, in _worker_loop
data = fetcher.fetch(index)
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
return self.collate_fn(data)
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 73, in default_collate
return {key: default_collate([d[key] for d in batch]) for key in elem}
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 73, in
return {key: default_collate([d[key] for d in batch]) for key in elem}
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 63, in default_collate
return default_collate([torch.as_tensor(b) for b in batch])
File "/home/pingan/miniconda3/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 55, in default_collate
return torch.stack(batch, 0, out=out)
RuntimeError: stack expects each tensor to be equal size, but got [8, 64000] at entry 0 and [8, 205052] at entry 7

Thanks for sharing this dataset!

I plan to train and evaluate pyannote speaker diarization pipelines on AISHELL-4.

1. I'd like to understand the speaker diarization labels better. In particular, I'd like to know if speaker labels are global to the whole dataset or only local to each file. For instance, can we assume that speaker 001-M in file 20200705_M_R002S01C01 is the same as speaker 001-M in file 20200616_M_R001S01C01? Or, are speaker labels recycled and inconsistent across files?

2. Are you aware of any published speaker diarization results on AISHELL-4?