This program is designed to superimpose an image mask on a waveform. Images should be drawn on a white (RGB(255, 255, 255)) canvas. An example image is included. Inspired by this YouTube video.
Requires numpy, pillow, pysoundfile, e.g. through:
pip install numpy Pillow PySoundfile
python3 image-to-waveform.py in/smile.png out/smile.wav
-i INAUDIO, --inAudio INAUDIO
Input audio filename
-d DURATION, --duration DURATION
Length of generated audio in seconds
-f FREQUENCY, --frequency FREQUENCY
Generator frequency
-s SAMPLERATE, --samplerate SAMPLERATE
Generator sample rate
--interp {nn,lin} Interpolation type
The program transforms each column of the image into DC offsets and amplitudes. The resulting parameters are then multiplied with the original waveform. Interpolation has to be done to smoothly convert from the x-axis of the image to the time-axis of the audio. As time only increases, the image also has to have certain properties in order to be fully representable in the audio (there should be no discontinuities in the y-axis).
The default waveform is a cosine wave at maximum amplitude with the Nyquist frequency. This frequency is chosen such that the resulting samples continuously alternate between -1 and +1, and thus provide most visible body to the resulting waveform. Of course, the user is free to choose any other frequency or even to load their own waveform. The default result is inaudible for the average human (22050 Hz), but should be readily visible through e.g. Audacity.