Real time system for extraction and playback of an instrumental sound

This paper presents an acquisition system for sound recording, typical spectral component identification and extraction of particular instruments from a soundtrack, based on each instrument's spectral components. This system is also capable of sound reproduction (playback) and its main feature consists of minimizing the delay between the recorded sound and the playback sound below 100ms. To achieve all these objectives, the proposed system consists of several components which will operate in parallel: the acquisition module which outputs the digital waveform of the sound and its FFT transform used for spectral representation; the comparison module, composed of associative modules which compares the closeness of the sound sample to certain templates which contain spectral representation of pre-recorded instruments, used to identify whether certain instrument is present; the spectral form extraction module which isolates a certain instrument from the original sound-mix, based on filtering the spectral components that don't belong to the respective instrument; the last module is the inverse FFT module, which converts the sound from spectral domain back to time domain. With the exception of the microphone, the analog-to-digital converter (ADC) and the digital-to-analog converter (DAC) capable of outputting the sound, all other operations are done by an SoC integrated system (Xilinx Zynq 7000).