Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Single-channel speech enhancement by using psychoacoustical model inspired fusion framework

Suman Samui

TL;DR

The paper tackles the quality-intelligibility trade-off in single-channel speech enhancement by merging a psychoacoustic-inspired acoustic-domain parametric Bayesian STSA estimator with a modulation-domain binary masking approach. It derives a closed-form gain under a generalized gamma prior with adaptive parameters, and introduces frequency-dependent α/β tuning alongside a dual-stage fusion in the STFT domain with an SNR-dependent weighting. Experimental results on Harvard sentences across multiple noise types and SNRs show consistent gains in PESQ and ESTOI over several baselines, demonstrating improved speech quality without sacrificing intelligibility. The work offers a practical fusion framework with potential benefits for robust noisy-speech applications.

Abstract

When the parameters of Bayesian Short-time Spectral Amplitude (STSA) estimator for speech enhancement are selected based on the characteristics of the human auditory system, the gain function of the estimator becomes more flexible. Although this type of estimator in acoustic domain is quite effective in reducing the back-ground noise at high frequencies, it produces more speech distortions, which make the high-frequency contents of the speech such as friciatives less perceptible in heavy noise conditions, resulting in intelligibility reduction. On the other hand, the speech enhancement scheme, which exploits the psychoacoustic evidence of frequency selectivity in the modulation domain, is found to be able to increase the intelligibility of noisy speech by a substantial amount, but also suffers from the temporal slurring problem due to its essential design constraint. In order to achieve the joint improvements in both the perceived speech quality and intelligibility, we proposed and investigated a fusion framework by combining the merits of acoustic and modulation domain approaches while avoiding their respective weaknesses. Objective measure evaluation shows that the proposed speech enhancement fusion framework can provide consistent improvements in the perceived speech quality and intelligibility across different SNR levels in various noise conditions, while compared to the other baseline techniques.

Single-channel speech enhancement by using psychoacoustical model inspired fusion framework

TL;DR

The paper tackles the quality-intelligibility trade-off in single-channel speech enhancement by merging a psychoacoustic-inspired acoustic-domain parametric Bayesian STSA estimator with a modulation-domain binary masking approach. It derives a closed-form gain under a generalized gamma prior with adaptive parameters, and introduces frequency-dependent α/β tuning alongside a dual-stage fusion in the STFT domain with an SNR-dependent weighting. Experimental results on Harvard sentences across multiple noise types and SNRs show consistent gains in PESQ and ESTOI over several baselines, demonstrating improved speech quality without sacrificing intelligibility. The work offers a practical fusion framework with potential benefits for robust noisy-speech applications.

Abstract

When the parameters of Bayesian Short-time Spectral Amplitude (STSA) estimator for speech enhancement are selected based on the characteristics of the human auditory system, the gain function of the estimator becomes more flexible. Although this type of estimator in acoustic domain is quite effective in reducing the back-ground noise at high frequencies, it produces more speech distortions, which make the high-frequency contents of the speech such as friciatives less perceptible in heavy noise conditions, resulting in intelligibility reduction. On the other hand, the speech enhancement scheme, which exploits the psychoacoustic evidence of frequency selectivity in the modulation domain, is found to be able to increase the intelligibility of noisy speech by a substantial amount, but also suffers from the temporal slurring problem due to its essential design constraint. In order to achieve the joint improvements in both the perceived speech quality and intelligibility, we proposed and investigated a fusion framework by combining the merits of acoustic and modulation domain approaches while avoiding their respective weaknesses. Objective measure evaluation shows that the proposed speech enhancement fusion framework can provide consistent improvements in the perceived speech quality and intelligibility across different SNR levels in various noise conditions, while compared to the other baseline techniques.
Paper Structure (11 sections, 14 equations, 3 figures, 2 tables)

This paper contains 11 sections, 14 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Signal model and notation
  3. Gain function ($G_A$) of the parametric Bayesian STSA estimator
  4. Derivation of gain function
  5. Selection of parameters: $\alpha$ and $\beta$
  6. Modulation domain processing
  7. Proposed fusion framework
  8. Experimental results
  9. Speech and noise corpus
  10. Objective measure evaluation
  11. Conclusions

Figures (3)

  • Figure 1: (a) Gain of the parametric STSA estimator $(20\log G)$ versus instantaneous SNR $(\gamma-1)$ for several values of $\beta$. (b) $20\log G$ versus $(\gamma-1)$ for several values of $\alpha$ (c) $20\log G$ versus $(\gamma-1)$ for several values of $\mu$ for $\zeta$ = 0 dB
  • Figure 2: $\alpha$ and $\beta$ variations with frequency
  • Figure 3: Magnitude spectrograms: (a) clean utterance (PESQ: 4.5) (b) speech degraded by babble boise at 0 dB SNR and noisy speech enhanced using (c) MMSE-LSA (d) PMB-STSA (e) MCS (f) the proposed fusion approach.