Sound Recognition

Using Neural Networks - Convolution2D

Technical specs

Python(Keras lib for Neural Networks) + OpenCV 3.0

Summary

The main goal of this project is to recognize (fingerprint) short audio samples, such as short speech command, whistle, or any other sound from nature and map them to specific action.

Motivation

Using sound samples for reaching your goal: - Sound recognition of songs (music) , sounds from nature, human voice - Using human voice for commanding smarthpone, smart vehicle during the ride - Can be of great use to people with major disability.

Implemented:

- So far, software is trained to recognize whistle melodies and short audio samples. 
It can be easly upgraded to recognize specific types of sound.

Further implemenation:

 - Sound recognition in real-time (not from audio samples, live recording from mic)
 - New data-sets and new training

Screenshots:

• Main frame:
  

Simple whistle ASCENDING test:

• Application output:
  
• Test analyze - FFT:
  
• Test analyze - Waveform:
  
• Test analyze - Spectrogram:
  
• Test analyze - Spectrogram - Black and White (ready as Neural Network input):
  

Instalation

Licence

- MIT

References

Great spectrogram article <br/ > University Of Novi Sad, Faculty Of Techical Scieces, AI-lab