I have an issue with the recorded data being stretched (oversampling) , I’ve run some tests and here’s how it looks. I have composed a record with repeated sin tones (10 secs each) at constant intervals of 10 secs totaling 250 secs, played it and recorded the signal on the matrix creator , at different rates, and on a smartphone as control
The issue is found at all rates (6k , 44.1k , 48k) using arecord with matrixio-kernel-modules.
I’ve also encountered this effect while using the matrix_HAL driver.
kindly find attached a screenshot from audacity and a link to the audacity project files.
Hello, I have a same issue.
We are talking about many packet losses of the data. As it looks, some data is “swallowed”. At sampling rate of 44100 Hz it happens (partially) in regular intervals (visible peaks with the interval of 1sec and missing parts of the wave).
On the other hand at a smaller sampling rate e.g. 16kHz no comparable peaks can be found. Nevertheless, the recordings are not of the same length as the source. So, decreasing the sampling rate brings some improvement (recording is steadier), but does not solve the problem completely. The positive thing is that the packet loss happens synchronously (with all 8 microphones/channels at the same time), so e.g. the direction finding is still possible.
On the screenshots, you can see the following waveform and spectrogram of the audio files:
- Generated test signal 120 sec long with 5 sec start and stop 1 kHz sine signal and a 30 sec 500 Hz sine signal in the middle.
- A1_1 recording with MEMs array, 120 sec, 5 channel
- A1_2 recording with MEMs array, 120 sec, 5 channel (same conditions as before)
- recording with MEMs array, 180 sec, 5 channels
at 16000 Hz:
- Generated test signal for 120 sec with 5 sec start and stop 1 kHz sine signal and a 30 sec 500 Hz sine signal in the middle.
- recording with MEMs array, 120 sec, 5 channel
- recording with MEMs array, 120 sec, 5 channel (same conditions as before)
If someone know a solution of the problem - please let me know!