Paul Guyot

Results 140 comments of Paul Guyot

Finally finished the tdnn_f model. %WER 25.40 [ 35749 / 140755, 3408 ins, 11742 del, 20599 sub ] exp/nnet3_chain/tdnn_f/decode_test/wer_7_0.0 Downloadable from the same link. https://github.com/pguyot/zamia-speech/releases/tag/20190930

You can find precompiled binaries for Raspberry Pi here: https://github.com/pguyot/kaldi/releases https://github.com/pguyot/py-kaldi-asr/releases

Most of the training (time-wise) does not actually require a GPU. The GPU is only used at the very end of the process, and the script aborts too early. I...

@joazoa Sorry if my previous message was unclear about CPU/GPU requirements. I have been renting a VM with a GPU and I found out that the GPU is required too...

So eventually we decided to put adc into the VM? Doesn't it imply adding an additional component dependency?

Could you please try #102 and let me know if it works for you?

Do you need to optimize the image afterwards to use it as an artifact? If not, you can try `optimize_image: false`. Likewise, I wonder if `boot_partition` option can help.

Thank you. I'm not sure it's a good idea to have this case in the CI of the action as it takes about 7 minutes as the image is so...

Nvidia images are large and GitHub runners have less available free space, so the test partitions was broken. I fixed it by deleting stuff...

Exactly. This can be achieved by: - using a container or runner that has an apt package with the qemu patch and installing them, so that: `apt install -y qemu...