TOPIC:

Hi,
We purchased 2 Mic Arrays.
We tested various configuration (beam, raw, ... with or without any setttings) and the audio is by far not as good as a regular microphone.

I am wondering whether this product plays in the same game as the chips in the Amazon Echo or such device ....

Any clue ?

vnguyen

Very odd statement of "very poor audio".. never heard of this before and haven't seen similar statements..
maybe you could provide more info? Did you maybe play/monkey with the settings? That could easily turn some issues.

Also, under "raw audio", you're looking at 8ch of unprocessed audio @ 16kHz. What's your way to evaluate audio?

Amazon echo is using similar PDM microphone quality, for voice activated application, indeed similar quality if you were to look at the feed of I2S going to the cloud.
More info, engineering statement as to the issue would help I think,..

DevTeam

Hi there,

I was conducting the experiments with vnugyen. The tests were simple : play a rather clean audio through rather good speakers and record with the mics. We did this with one UMA-8 in directional bf mode (#1) and the other in DOA mode (#2). All the other settings were kept to default (see screenshots). As a reference we used an audio technica cardioid 44 pro to record in the same set up.






The difference to the ear is quite significant thus the 'very poor audio' statement. Please listen to the samples here: goo.gl/ur65r9

F

Comparing UMA-8 against a pro-level (low noise) cardioid pointed at the talker in a controlled environment is a little unfair as the intended use cases are different. The pro cardioid is intended to sound great (flat response, low-distortion) and minimize a small amount of room reflections (4.8dB is the Directivity Index, or DI) in a recording studio environment. The UMA-8, following Echo and like devices have to deal with unknown user directions, interfering talkers, room configurations and ultimately are geared toward passing audio to a speech-to-text algorithm. The emphasis is therefore on rejecting acoustic noises in the vicinity via beamforming directly to a user's mouth and AEC of the device output path. If bandwidth, flatness and some added broadband (white) noise gain are the result, that matters relatively little to ASR algorithms. Which is why a 16kHz sample rate for 8kHz bandwidth on the chip is typically acceptable. Hi-fi, certainly not. But intelligible, yes. Distortion and time-shifting effects like jitter do tend to be a problem for ASR, to the best of my knowledge.

So...your recordings of the UMA-8 boards sound "poorly normal" to my ears for a multi-mic setup with DSP on.

That being said, there could always be areas for improvement. Turning AEC off might help. Typically this has a negative impact on audio quality. Some physical acoustic suggestions involve keeping your array close to the table and away from walls to minimize offensive in-band "comb filtering" from reflections. Keep the mic ports open to the air and with no diffraction from a housing or grill. Also (ahem... miniDSP developers), the mic-to-center mic spacing of around 45mm on the UMA-8 is on the large side. A cardioid by subtracting two mics at this spacing will have a null at 3.8kHz (fnull,cardioid = c/2d) where c = 343 m/s and d = 45mm. Larger spacing means more sensitivity (yay!) at the cost of bandwidth (boo!)- so there is a balance point. However, generally one would target somewhere around the Nyquist frequency of the codec (8kHz), so a board to about half the diameter of the current board would likely have better audio quality.

Dear all

Great email already from marcpbcup (hard to write that name.. !
It's spot on... Remember that the UMA-8 is intended for voice activated applications. i..e far field microphone.

One comment worth adding is maybe the mic was upside down when doing the recording? That's the typical raised noise floor you'd hear. LED are supposed to be on the bottom like that pic (www.minidsp.com/applications/usb-mic-arr...-rpi-diy-amazon-echo)