akul2010 / sounddevice_vad Goto Github PK

The purpose of this project is to come up with a more streamlined audio system for a voice assistant. The goal is to demonstrate the system listening, thinking and talking all at the same time through the use of multiple threads.

I wanted to try using the sounddevice library rather than objects derived from the audioengine classes in Naomi. The ALSA class doesn't work for mixing streams, and the pyAudio class has trouble identifying devices. They are both somewhat problematic and prone to stuttering. Sounddevice uses pyAudio and is actually pretty similar to the AudioEngine objects Naomi has been using.

I'm still evaluating whether I can add sounddevice as a new option and continue using the pyAudio and Alsa AudioEngine objects, or if this rewrite requires retiring those options. If anyone has a preference, please let me know.

I am just letting the sounddevice use its own default device rather than specifying a specific device. Sounddevice claims that this will be what you want most of the time. It seems to work. I would require the user to choose a device during the actual Naomi setup, so I am looking at a list of compatible devices that could be presented to the user to select from, the same we are now. I can't seem to redirect the error messages generated while the devices are being tested, though.