search

SpeechColab / Leaderboard

SpeechIO Leaderboard: a large, robust, comprehensive, benchmarking platform for Automatic Speech Recognition.
434 stars 61 forks source link

Request to test our model trained with private data. #16

Closed iou2much closed 2 years ago

iou2much commented 2 years ago

Hello. Jiayu. We request to test our model trained with our business data.

As there's no public API yet, we encrypt the model and upload to the OSS. And it could only be used with the asr binary package, which you can find in Dockerfile.

iou2much commented 2 years ago

BTW, sorry for that I accidentally push a coredump file to the OSS. As I thought it only uploads assets folder, and forgot to clean up first. Please delete the core.* file in OSS to avoid costing extra fees.

dophist commented 2 years ago

There seems to be some permission problems when i was reproducing your model-image: 1). scripts in assets/ don't have execution permission, and top level scripts call them in a ./xxx style, I fixed this manually by chmod +x assets/xxx 2). after fixing 1), log.SBI gives: /sbin/ldconfig.real: Can't link /opt/libtorch/lib/libgomp.so.1 to libgomp-52f2fd74.so.1 /sbin/ldconfig.real: Can't create temporary cache file /etc/ld.so.cache~: Permission denied

I haven't got time to look into your docker env for more details, maybe you can validate first on MINI or ZH00000(or possibly other unlocked sets), to see any of these happens.

iou2much commented 2 years ago

Thank you for looking into this PR. Actually I tested the whole procedures on my server , and I haven't seen any of these errors.

I think the 2nd issue might be related to the user permission, as I use root user on my server, and you might be using normal user ? Never mind. I change the SBI. it no long asks for super permission now. Please try again. Thanks.

This is my result on opensource datasets: image

iou2much commented 2 years ago

And please pull the model again, thank you.

dophist commented 2 years ago

Cool, now it runs smoothly. Merging.