Real time system for instrumental sound extraction and recognition

摘要

This article presents a “smart sound sensor” solution. More specifically it is a system on a chip capable of extracting sound patterns from an input signal captured by a microphone, patterns that correspond to certain musical instruments. The aim is to identify sounds like: voices, ambient sounds or musical sounds. Such applications generally involve the presence of a computer that analyzes offline the audio sequence. There are also smartphone security applications that rely on voice recognition templates that were previously stored and which work by processing the input sequence offline, after recording. Unlike these methods, our solution enables online real time identification of a group of musical instruments and allows extracting of certain sound sequences that belong to a particular instrument so that it can be isolated for sole playing. The solution was implemented in an integrated system “sound sensor” that analyzes the input signal in real time and can give output data without additional elements.