Purpose

Run knn on audio data which has been converted to images to find similar audio clips

Ffmpeg preporcessing command to create spectrogram image

Spectogramspectrogram

for %f in (*.wav) do ffmpeg -y -i "%f" -ar 44100 -lavfi "showspectrumpic=s=256x256:scale=log:legend=0" "img\%~nf.png"

With loudnorm to try to improve performance by normalizing loudness

^{Seems to work much better}

for %f in (*.wav) do ffmpeg -y -i "%f" -ar 44100 -lavfi "loudnorm,showspectrumpic=s=256x256:scale=log:legend=0" "img\%~nf.png"

stb_image.h

Used for gathering the image data

Building

clang src/main.c -Iexternal -Lexternal -lOpenCL -O3 -fopenmp -march=native -o soundknn.exe

Todo

• Load all images from directory
• Do knn algorithm
• Use opencl to accelerate algorithm execution