njusq

Results 3 comments of njusq

我也碰到了同样的问题(虽然不是同一个程序),我正在用tfdbg调试,能帮助查找程序中出现的nan值,后来发现是自己之前没注意的一个地方存在0除以0导致了nan的出现。希望对你有帮助。

We also had the same problem two days ago under the setting `subtrach_mean = False`. We compared the results of torchaudio's fbank and kaldi's compute-fbank-feats line-by-line. The differences occured from...

同发现了CPU推理速度很慢,但我测出来的RTF大概在1.19左右,没有你的结果那么。请问后来你有找到改善方法吗?