search

NVIDIA / NeMo

A scalable generative AI framework built for researchers and developers working on Large Language Models, Multimodal, and Speech AI (Automatic Speech Recognition and Text-to-Speech)
https://docs.nvidia.com/nemo-framework/user-guide/latest/overview.html
Apache License 2.0
11.84k stars 2.46k forks source link

Quartznet inference not deterministic #1030

Closed janvainer closed 4 years ago

janvainer commented 4 years ago

Hi, I am running quartznet inference with the inference script from the docs and for a single recording, I am getting slightly different result every time. I noticed because the WER is slightly different on every run. Why is that and how to make the run deterministic?

okuchaiev commented 4 years ago

In pre-processor params (https://github.com/NVIDIA/NeMo/blob/master/examples/asr/configs/quartznet15x5.yaml) there is a non-zero "dither" which introduces some randomness into the signal.

janvainer commented 4 years ago

ok thanks :)