Aalto Acoustics Lab

Audio Processing Using Giant FFT

The fast Fourier transform, FFT, computes quickly and accurately the complex spectrum of very long signals containing several minutes or hours of audio. This capability has enabled new audio processing methods based on modifying a signal’s global spectrum.

Description

The ability to compute the FFT of an entire audio signal and modify its spectral properties has led to a new family of methods known as giant FFTs. Inspired by this idea, Vesa Välimäki and Stefan Bilbao introduced an FFT-based sample-rate conversion technique in 2023 that is both fast and accurate. A phase-randomization technique combined with zero-padding extends a short audio sample into a static soundscape of arbitrary length, also called an endless sound. Other giant-FFT applications include the generation of sonic palindromes and babble noise using zero-phase processing, as well as pseudo-time-stretching achieved by scaling the global phase by a constant.