search

rolczynski / Automatic-Speech-Recognition

🎧 Automatic Speech Recognition: DeepSpeech & Seq2Seq (TensorFlow)
GNU Affero General Public License v3.0
223 stars 62 forks source link

New features #30

Closed qbit- closed 4 years ago

qbit- commented 4 years ago

Feature extractor has been rewritten to coincide with nvidia nemo. The interface of the following was changed: FeatureExtractor utils.read_audio

rolczynski commented 4 years ago

hey folks @ValeriyMarkin @qbit- @RunFMe ! I am super glad that you use this package. I wonder maybe do you wish to make a valid PR to the master hm?

qbit- commented 4 years ago

Hi Rafal, thanks for a great code! We chose your project because it is simple and modular. Our stuff is currently very much WIP and poorly tested, and many things are broken. We would be glad to PR after the master branch gets cleaner. By the way, we mostly did the following: 1) implemented a new feature extraction part 2) added a couple of models 3) used horovod parallelization instead of keras MirroredStrategy

rolczynski commented 4 years ago

Great! Please do not forget about this PR 👍
Of course, I can help you somehow.