Frequency-Domain SNR tools in MATLAB®

Code demonstrates the Frequency-Domain SNR tools described in [1].

• Four source audio signals x are provided in the audio directory, as are three degraded versions xhat of each source (16 .wav files total).

• Demo.m reports 7 FD SNRs for xhat relative to x. It then performs simulated resoration of the choosen type and level and reports the resulting 7 FD SNRs. Finally, it peforms Griffin-Lim of the choosen type and number of iterations and again reports the 7 FD SNRs.

• Note that Scale-invariant SNR is provded as a simple and easy-to-distribute tool that compares the resulting time-domain signals but it is not a good proxy for speech quaility. For any serious work it would be good to aquire and apply one or more tools that better track human perception (e.g., PESQ, POLQA, ViSQOL, STOI, eSTOI).

If this code is useful for your research, we would appreciate if any resulting publications would cite [1].

[1] S. Voran and J. Pieper, "Frequency-Domain Signal-to-Noise Ratios Illuminate the Effects of the Spectral Consistency Constraint and Griffin-Lim Algorithms," 2025 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, Tahoe City, CA, USA,Oct. 12-15, 2025.

How to run

• Clone to local repository
• Start MATLAB or Octave
• Run demo.m
• Follow comments inside demo.m to modify input values for additional demos
• Output values are displayed with labels, but could be stored with a bit more code