Welcome to Phase Processing Lab conducted by Pejman Mowlaee CV
Harmonic Phase Estimation in Single-Channel Speech Enhancement Using Phase Decomposition and SNR Information
- Audio samples -
We propose harmonic phase estimator for single-channel speech enhancement. The so-obtained phase estimates can be used directly for signal reconstruction or be combined with an amplitude-enhancement only as in MMSE-LSA.
P. Mowlaee, J. Kulmer,"Harmonic Phase Estimation in Single-Channel Speech Enhancement Using Phase Decomposition and SNR Information: listening examples", submitted to IEEE Trans. Audio Speech and Language Processing, June 2014.
- Part 1: Phase enhancement (PE) combined with amplitude enhancement (LSA) -
Male speech corrupted in white noise at SNR = 5 (dB):
|
|
|
|
|
Male speech corrupted in white noise at SNR = 10 (dB):
|
|
|
|
|
Female speech corrupted in babble noise at SNR = 5 (dB):
|
|
|
|
|
Female speech corrupted in babble noise at SNR = 10 (dB):
|
|
|
|
|
|
- Part 2: Phase-only enhancement + noisy spectral amplitude -
- SNR= 10 (dB) Babble Noise -
Here we present some audio examples for phase-only enhancement scenario with PESQ and STOI results:PESQ Scores: Noisy = 1.93, phase-enhanced = 2.27.
STOI Scores: Noisy = 0.78, phase-enhanced = 0.83.
|
|
|
|
|
|
- Part 3: Phase enhancement + LSA spectral amplitude -
- SNR= 10 (dB) Babble Noise -
Here we present some audio examples for phase-only enhancement scenario with PESQ and STOI results:PESQ Scores: Noisy = 1.93, LSA amplitude = 2.16, phase-enhanced = 2.48.
STOI Scores: Noisy = 0.78, LSA amplitude = 0.77, phase-enhanced = 0.81.
|
|
|
|
|
|
|
|