Inversion of Magnitude Spectrograms with Adaptive Window Lenghts

Presented at the 34th International Conference of Audio, Speech, and Signal Processing (ICASSP-09), April 19-24, 2009, Taipei, Taiwan (ROC).

Conference homepage

Abstract

In this paper, we extend the Real-Time Iterative Spectrogram Inversion method (RTISI) for generating a time-domain audio signal from a magnitude spectrogram such that it can handle changing spectrogram window lengths. For each desired window length, we use a separate buffer structure and synchronize the buffers each time the window length changes. This way, the proposed method helps to improve the time/frequency-resolution trade-off for algorithms that operate on magnitude-only spectra.

Paper

PDF

Poster

PDF (2 MB)

Sound Examples

Mix of a castanets and double bass signal. Original sources: EBU-SQAM

All sound examples are stored in the FLAC format (Free Lossless Audio Coding). Details and decoding software can be found here.

Original Single-resolution Phase Estimation Multi-Res. Phase Estim.
512 Samples 1024 Samples 2048 Samples 512/2048 Samples
FLAC FLAC FLAC FLAC FLAC
Note that all of these sound examples contain already the improvements presented in this paper.