I want to detect the direction of sounds (for a social robot), and recognise speech in a noisy environment. I wondered if anybody knows of any good solutions.
I have used the ReSpeaker Mic Array 2.0 in the past with some success, but I’m hoping for something better or cheaper.
The ReSpeaker board can provide DOA, but although it contains a mic array, it is not really useful for beam forming or improving the signal to noise ratio of detected speech in a noisy environment. Also it seems to be a dead project. My conclusion after using it is that I could get just as good speech recognition with a cheap omnidirectional microphone. However the direction of arrival is really good, so is there an alternative that offers that?
Has anybody had success detecting sound direction with their own hardware? For example, I wonder if 2 or 3 microphones with some mechanical directional shield (i.e. artificial pinnae) could work well.
I have no experience with the UMA board. Both ReSpeaker and UMA are discussed in this thread. Neither get totally positive opinions. The ReSpeaker board seems to be a dead product (no updates to the code on github a long time). The UMA product IS very similar in hardware to ReSpeaker, but doesn’t seem to work very well.
The conclusion seems to be that there is no off-the-shelf product that can compete with Alexa and Echo for far-field speech recognition.