Dereverberation of audio having reverberation (echos) to improve ASR WER performance

We have a couple of audio files with quite a bit of reverberation. We would like you to use any deverberation ML algorithm that you know of to help improve the WER of our ASR decoder. We are using an end-to-end ASR, which current outputs a baseline WER. If you can improve the WER of our test files over our baseline WER by 5%, we will pay your $700; lift of 6% $800, lift of 7% $900, and lift of 10% - $1500. Results MUST be repeatable over several files.

what we will provide:
a) API to our ASR decoder
b) sample audio files with reverberation
c) ground truth of the audio files
d) script to benchmark the ASR output of your modified audio file against the ground truth.

Deliverables:
a) deverberation code you used that can achieve the 5+% lift.
b) instruction on how to use the code
About the recuiter
Member since Mar 14, 2020
Varsha Somaiya
from Parana, Brazil

---

---

Looking for help? Checkout our video tutorial
How to search and apply for jobs

How to apply? Do you have more questions about the Job?
See frequently asked questions