search

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP
https://nvidia.github.io/OpenSeq2Seq
Apache License 2.0
1.54k stars 371 forks source link

Integrate a pre-trained Speech2Vec embedding layer for Speech To Text models #423

Closed jsvir closed 5 years ago

jsvir commented 5 years ago
  1. Train an embedding using all audio data in the train set - optional setting in config file
  2. Use it in the data layer instead of audio features only

Based on the paper: https://arxiv.org/abs/1803.08976