NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP
https://nvidia.github.io/OpenSeq2Seq
Apache License 2.0
1.54k stars 369 forks source link

Add more feature extraction/data layer options #368

Closed ryanleary closed 5 years ago

ryanleary commented 5 years ago

This introduces an optional min_duration filter, in addition to making it possible to add dithering, specify desired window, and normalization type. Further adds ability to do speed perturbation without introducing additional noise, and prevents crashes due to divide by zero/asserts.

blisc commented 5 years ago

Merged as part of #399