mins0o / pcmlabeler Goto Github PK

I created a tool to measure and label audio PCM data for signal processing (Fourier transform etc.) & machine learning.
I started with Arduino AVR devices but realized their limitations of computing power and frequency.
Human audible frequency is considered to be around 20 Hz to 20 kHz, and initial method with an arduino I could only achieve 300 Hz. This was greatly improved after several experiments.

(Without any floating point calculation, Arduino's 2-byte integer arithmetic is very powerful. With a single floating point variable per loop, 11.18 kHz goes down to 1 kHz.)

These results are not accurate but an approximation. The results were consistant during the experiment, but it may vary on devices, and small change in code may alter the result dramatically.

The data is labeled as

• t : tap on mic
• c : clap
• w : whistle
• s : snap

• Examples
  

In this project, I learned

• how to use an Arduino as an USB-SerialTTL converter and how MCU programming works.
• how sampling rates work and how to measure it in MCU.
• comparing the cost-benefit of using MCU's memory or communication capability. The two different methods of record the audio PCM has their pros and cons.
• utilizing MCU's memory capacity to its max to increase sampling rate.
• performing orthodox(not FFT algorithm) Fourier transform on PCM data.