Directional microphone about diy-alexa HOT 5 CLOSED

I tested MASE. It's just a noise reduction algorithm, it has nothing to do with beamforming. At the same time, it does not work well - it greatly distorts the sound and cuts the frequencies to 4 kHz. In the future, the automatic voice recognition system recognizes the voice much worse than without MUSE.

from diy-alexa.

I think you could definitely do it at sample capture time - as you say - before putting the samples into the ring buffer.

If memory does become an issue then I'd be tempted to look at the wrover modules which have 4MB of RAM.

from diy-alexa.

Yeah my wondering mind keeps thinking that a distributed array of directional mics that use the KW hit result could give a central ASR a signal to use the best and nearest mic signal.
But yeah the wrover dont have much price difference now.
https://www.ebay.co.uk/itm/ESP32-WROVER-B-T8-V1-8-ESP32-8MB-PSRAM-TF-Card-WiFi-Module-Bluetooth-Board-HL/293784167596
Would need 2x i2s mics but still very cheap.

Does anyone know if there is a lib for a lossless audio RTP with latency compensation similar to snapcast?

from diy-alexa.

Whilst we are talking mics has anyone tried the MASE and other algs on Wrover or lower?
https://github.com/espressif/esp-sr

If it was a RTP mic streamer triggered by KW till silence that would be amazing to couple to a central ASR.
If it maybe streamed AMR-WB to a port and accompanied KWS hit score on another then multiple ESP32 could be part of a distributed array.

from diy-alexa.

@andrej102 I did wonder how they had shoehorned a series of load heavy DSP that even seems to be too much for Raspberry embedded Linux. 4kHz is no use to the central ASR I plan to use.

They have done an amazing demo with the esp32-s3 https://www.youtube.com/watch?v=ARkuaeEW1eM

The EQ DSP to rectify the filtering of the beamforming also adds much load that the stereo mic idea I gave up on.
The ADC on the ESP32 from bits to SNR is not great but going analogue with directional electrets is an alternative.

The above vid has me confused as it seems to exceed high end dsp beamformers I have seen and if snakeoil or not I am unsure.
I still plan to use distributed directional electrets so that a unit is voice=near / noise=far purely by positioning

from diy-alexa.