Why use argmin(tau) for DoA on 6-mic array?

[ruihuili]

Dear Voice Engine developers,

I find your voice engine module for KWS and DOA detections particularly useful, so first of all, thank you for providing this important building block for many voice related applications.

While I am reading the code for the 6-mic circular array DOA, I found in line 45 of doa_respeaker_v2_6mic_array.py:

min_index = np.argmin(np.abs(tau))

I was wondering why you take only one pair of the mics rather than 6 mics all together to compute the angle of arrival? and why specifically the one pair with the minimum tau is taken to compute the 'best_guess'?

It would be much appreciated if you could kindly elaborate on this.

All the best
Rui

[xiongyihui]

因为1对麦克风比用6个，计算量要小很多，实现也简单多了。
用最小角度的1对麦克风是因为角度小的时候分辨率高一些（垂直与2个麦克风方向，分辨率最高）。

[ruihuili]

I see. Thank you for your reply :)

The other question I have is, when I plot all possible theta (phase shift), it seems only 9 values (of phase shift) for each microphone pair is considered for cross correlation and these values are centred around 0. Is this also due to computational power? But this could result in inaccuracies in angle of arrivals I suppose?

I am also wondering on which pi version the algorithm is developped? because I was thinking perhaps with newer versions of Pi the computational constraint at least is more relaxed?

Thank you in advance for any info

Best,
Rui

[xiongyihui]

如果想把6个麦克风都用上，常见的使用SRP-PHAT搜索平面360°，半球面。
除了pi 2+还可以都可以算的过来的。可以了解这个项目 https://github.com/introlab/odas