A proposed method for the sound recognition process

Abstract EN

One of the most important issues in the signal-processing world is the issue of sounds differentiations.

It has many applications in identifying the sources of the sounds, and many researches nowadays serves in this field and all of them looking for the best way to have a high accuracy implementation of the discrimination process.

Sounds can be recognized under suitable recording conditions by converting the sound signal from time to frequency domain (because Sound signal of a source differs from other sources by the frequency contents.

This property serves to differentiate between sounds, and differences become visually apparent when the spectrograms of the signals are compared).

All the classical methods based on the amplitude comparison of the spectrum.

The main problem faces the recognition process is the 100% system accuracy cannot be achieved.

A proposed strategy suggested solving the problem.

It is based on the comparison of slops between spectrogram sections instead of the magnitude comparison and taking the minimum differences between the pattern and references stored in a database.

The tested examples with the help of Matlab program proved that the proposed method is more accurate than the conventional methods.