spiralanch / deesser

Experimental university project. Audio processing program to get rid of excessive prominence of sibilant consonants

DeEsser

Audio processing for getting rid of excessive prominence of sibilant consonants.

Algorithm

Algorithm searches for excessive sibilants in audio by counting number of zero crossing in each area and apply filter on areas, where counted number is greater than certain constant. Filter itself is converting each area by Fast Fourier Transform, then suppress intensity of frequincies above certain threshold and converting signal back by Inverse Fast Fouriers Transform.

Build

git clone --recursive https://github.com/crataegus27/DeEsser.git
cd DeEsser
mkdir build
cd build 
cmake ..
make # your build command dependent on platform

Dependencies

• nanogui
• libsoundio
• fmt
• stb_vorbis
• vorbis + ogg

Progress

1. Project initialization (cmake)
2. Loading and decoding ogg audio file from disk (stb_vorbis)
3. Streaming audio data to speakers through (libsoundio)
4. Rendering data like waveform, audio specs, playback control (nanogui)
5. Audio processing, fft -> suppress frequencies -> ifft
6. By average peeks finding
7. Interface to tune settings
8. Saving to ogg file (vorbis + ogg codec)