Speech Signal Processing

Perceptual Evaluation of Phase-Aware Speech Enhancement in the Short-Time Fourier Transform Domain



Audio examples

Below you find some examples of the audio files that have been used for the perceptual evaluation of phase-aware speech enhancement presented in the open access publication

Martin Krawczyk-Becker, Timo Gerkmann,"Perceptive Evaluation of Phase-Aware Speech Enhancement in the Short-Time Fourier Transform Domain", submitted as an Express Letter to the Journal of the Acoustic Society of America.


The speech signals are taken from the TIMIT database, while the babble as well as the pink noise come from the NOISEX-92 database. To introduce some impulsive noise, we further added some hammer blows to the babble noise signal. Finally, we also tested a recording of a stroll in a park.

PBA (Phase-Blind Amplitude estimation) denotes Ephraim and Malah's short-time spectral amplitude estimator (STSA), which serves as a conventional, phase-blind reference. The estimated amplitude is combined with the noisy phase.

PAA (Phase-Aware Amplitude estimation) denotes the phase-aware amplitude estimator presented here. The estimated amplitude is combined with the noisy phase.

PAC (Phase-Aware estimation of Complex coefficients) denotes the case that the phase-aware amplitude estimate of PAA is combined with the clean speech phase. Please see the paper for details.

Babble noise at -5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)

Babble noise at 5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)

Stroll in a park at -5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)

Stroll in a park at 5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)

Pink noise at -5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)

Pink noise at 5 dB SNR:

Clean Speech Noisy Speech
PBA PAA (using exact clean phase)
PAC (using exact clean phase) PAA (using estimated clean phase)