search

introlab / odas

ODAS: Open embeddeD Audition System
MIT License
780 stars 248 forks source link

Respeaker 4-mic array: LED is way more accurate than SSL #213

Open ruofeidu opened 3 years ago

ruofeidu commented 3 years ago

When observing the raw data from SSL and the live results from odas_web, I found that the LED indicator is way more accurate than then the visualization and the raw data. Sometimes I saw one or two sound source potentials keep remaining in the visualization and raw data, even if the source source is no longer speaking.

So my questions are:

  1. How can I replicate the accurate algorithm of the LED indicators on the 4-mic array board?
  2. How can I best process raw SSL data to get the dominant sound source? I found the energy is also larger than 0.998 even though sometimes the sound source stops speaking.
jake266637 commented 3 years ago

Hi, I'm pretty sure the reason for the high accuracy of the LED indicators is because they do not track sound sources and rather dispaly instantaneous (every 8ms) DOA information. The tracking function in my experience is far less accurate. If you pull the json file (see image) which is found in the configuration file, this is the information that is fed into the LED indicators. So it is available but does not have any audio stream asociated with it. image