The past couple weeks have consisted of a series of small experiments aimed at testing the capabilities of the ReSpeaker microphone array and the Raspberry Pi. The process and results from these experiments are briefly described below.

Frequency Response

For this test, a small, portable Bluetooth speaker was used as a sound source and an online tone generator created the sinusoidal waveforms. The frequencies tested were: 60Hz, 120Hz, 240Hz, 480Hz, 960Hz, 1.4kHz, 10kHz, 16kHz, 19kHz, and a control sample with no tone playing. Each frequency was recorded at 48kHz sampling rate in 32bit float format.

Exposition: To ensure the speaker was capable of producing sine waves at these frequencies, audio produced by it was recorded by a Blue Snowball desktop microphone.

Physical Setup: The ReSpeaker, attached to the Raspberry Pi, was resting face-up on a pillow (to aid acoustic dampening) in the middle of a 12ft by 12ft room. The speaker was held by hand at a two-foot distance directly above the microphone array.

Procedure:

1. A tone frequency is selected and played through the speaker
2. The speaker was held above the microphone array
3. Microphone array is set to record
4. Record for 5-10 seconds, then stop the recording
5. Change frequencies and repeat.

Outcome: Analysis of each recorded frequency sample shows the device is capable of receiving audio frequencies within our desired range. Here is the spectrogram of the 60Hz and 19kHz frequencies.

In both spectrograms the specified frequency is clearly defined and the largest signal recorded by the microphone. The gap visible at 1.5s and 5s for the 1.4kHz and 19kHz graphs, respectively, are artificially generated by the online tone generator. Also visible are harmonics at regular intervals, for both spectrograms.

Future Further Testing: Further testing involving cross-checking with different speakers will be required to ensure that the harmonics are not issues with the microphone array.

Distance Range

This test was a bit more straight forward. It’s purpose was to test the range at which the device could capture audio samples clearly, and provide clear enough audio to work with the Direction of Arrival algorithm coming soon. Recordings were made when this person was standing at 10, 20, 30, 40, and 50ft. Each sample was recorded at 48kHz sampling rate in 32bit float format.

Physical Setup: The microphone array was placed on a pillow (to aid acoustic dampening) in a un-enclosed yard outdoors.

Procedure:

1. Standing at a specified distance, a person made clapping sounds at fairly constant volume
2. Microphone array is set to record
3. Record 8 to 12 clapping sounds
4. Change distances and repeat

Outcome: Below is an un-retouched clip of a waveform showing four claps at 40ft. Listening to the recording and visually inspecting show the device is capable of picking up audio from at least 40 feet.

Interestingly, the smaller spike after each initial spike (the clap) is from an echo bouncing off a wall several feet behind the microphone array.

Future Further Testing: Further testing needs to be done to check different frequencies at each distance, and at what volume they can longer be heard. This will help determine the physical distance the device can be from an identifiable sound source for an accurate prediction of direction and classification.