SeanNaren / deepspeech.torch

Speech Recognition using DeepSpeech2 network and the CTC activation function.
MIT License
260 stars 73 forks source link

Questions about training AN4 dataset #66

Closed Yizhen-Z closed 7 years ago

Yizhen-Z commented 7 years ago

I followed the installation guide and tried to run Train.lua on AN4 dataset. But the results I got was quite strange, the Average Loss turned out to be a huge number, while WER and CER were also larger than 100. (In my understanding, it should be a percentage number which is less than 100). screenshot

I didn't make any change on code. So I am really confused about it...

Another question is that, from the DS2 paper CER refers to 'Characer Error Rate' for the Mandarin system, so what is CER here refers to?

Thanks for any advise!

SeanNaren commented 7 years ago

Could you let me know what OS/GPU you are using?

Yizhen-Z commented 7 years ago

The OS is Ubuntu 16.04.1. GPU is NVIDIA GTX 1080.

Thanks for reply!

SeanNaren commented 7 years ago

Yeah I had the same issue as you, you need to install warp-ctc a bit differently on pascal cards, check here and try installing this branch instead, and restarting training!

Yizhen-Z commented 7 years ago

Yes, it works now. I finally got WER = 18, and CER = 3.88 on AN4 dataset. Thank you! loss

SeanNaren commented 7 years ago

Great stuff!