Voice Recognition System
Client company spent heavily to process audio in the search for the necessary intervals with the right sounds. Acute problem was the inefficient usage of the professionals in the state. This has caused a need for an automated searching process of right fragments in audios.
To reduce the processing time it was proposed to compress your original audio files using wavelet transformation, and to use modified similarity measures to search for designed templates.
This allowed the system to find the appropriate sound bites in long records (tens of millions of samples) of which the system did not known in advance.
The system handles 24 hour audio file in 3-5 minutes, and provides a time stamp on the entire file for the fragment, which was to be found. This allows you to increase the amount of information being processed, and thus the profit of the company
A program package was created for fast search of specific audio elements that were compressed based on wavelet transformation
A program package was created to compare and find the audio sample with the desired sound.
This package provides an entire list of elements similar to the specified one with the calculated degree of reliability (99.8%)