search

nishantgurunath / Separabl

2 stars 1 forks source link

Pretrained models #1

Open sberryman opened 4 years ago

sberryman commented 4 years ago

Any chance you are going to share pretrained models?

sberryman commented 4 years ago

I hadn't heard anything but did see your recent updates and have trained 5-10 models using 16 kHz and 24 kHz audio samples. I used my own dataset and managed to get very comparable results.

The muffled sound of the speaker is very apparent but it does remove the vast majority of background noise.

Have you looked into 2Hz.ai and their Krisp noise suppression? https://krisp.ai/

nishantgurunath commented 4 years ago

We are planning to share pretrained models pretty soon. Working on model generalization for different types of speech and background music such as Motivational/ Spiritual Videos, News Broadcast, How to videos etc.