Speech enhancement using a minimum mean-square error short-time spectral modulation magnitude estimator

Paliwal, Kuldip K.; Schwerin, Belinda; Wójcicki, Kamil

doi:10.1016/j.specom.2011.09.003

Cited by 64 publications

(44 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The performance of the minimum mean square error (MMSE) [7] and log-spectral amplitude MMSE (Log-MMSE) estimators [8] still remains among the best of the published methods [20]. In part, this can be attributed to their decision-directed estimation approach, which bases the spectral estimate of each frame partially on the estimates from previous frames via the a priori SNR estimate updated by using a memory coe cient [5].…”

Section: Previous Workmentioning

confidence: 99%

“…In part, this can be attributed to their decision-directed estimation approach, which bases the spectral estimate of each frame partially on the estimates from previous frames via the a priori SNR estimate updated by using a memory coe cient [5]. Spectral subtraction [21] and the decision-directed MMSE [20] methods have also been applied in the spectral modulation domain in order to better handle nonstationary noise.…”

Section: Previous Workmentioning

confidence: 99%

See 1 more Smart Citation

Spectral and Cepstral Audio Noise Reduction Techniques in Speech Emotion Recognition

Pohjalainen

Ringeval

Schuller

2016

Proceedings of the 24th ACM International Conference on Multimedia

View full text Add to dashboard Cite

Signal noise reduction can improve the performance of machine learning systems dealing with time signals such as audio. Real-life applicability of these recognition technologies requires the system to uphold its performance level in variable, challenging conditions such as noisy environments. In this contribution, we investigate audio signal denoising methods in cepstral and log-spectral domains and compare them with common implementations of standard techniques. The di↵erent approaches are first compared generally using averaged acoustic distance metrics. They are then applied to automatic recognition of spontaneous and natural emotions under simulated smartphone-recorded noisy conditions. Emotion recognition is implemented as support vector regression for continuous-valued prediction of arousal and valence on a realistic multimodal database. In the experiments, the proposed methods are found to generally outperform standard noise reduction algorithms.

show abstract

Section: Previous Workmentioning

confidence: 99%

Section: Previous Workmentioning

confidence: 99%

Spectral and Cepstral Audio Noise Reduction Techniques in Speech Emotion Recognition

Pohjalainen

Ringeval

Schuller

2016

Proceedings of the 24th ACM International Conference on Multimedia

View full text Add to dashboard Cite

show abstract

“…Recent research has shown that extension of this framework into the modulation domain may result in improved noise suppression and better speech quality [27][28][29]. Modulation frequency domain refers to another layer of frequency representation obtained by applying the STFT a second time (but with longer frames and frame advance) on the spectral amplitudes of a speech signal.…”

Section: Modulation Domain Processingmentioning

confidence: 99%

Speech enhancement in modulation domain using codebook-based speech and noise estimation

Mani

Champagne

Zhu

2015

2015 IEEE Global Conference on Signal and Information Processing (GlobalSIP)

View full text Add to dashboard Cite

Conventional single-channel speech enhancement methods implement the analysis-modification-synthesis (AMS) framework in the acoustic frequency domain. Recently, it has been shown that the extension of this framework to the modulation domain may result in better noise suppression. However, this conclusion has been reached by relying on a minimum statistics approach for the required noise power spectral density (PSD) estimation.Various noise estimation algorithms have been proposed over the years in the speech and audio processing literature. Among these, the widely used minimum statistics approach is known to introduce a time frame lag in the estimated noise spectrum. This can lead to highly inaccurate PSD estimates when the noise behaviour rapidly changes with time, i.e., non-stationary noise. Speech enhancement methods which employ these inaccurate noise PSD estimates tend to perform poorly in the noise suppression task, and in worst cases, may end up deteriorating the noisy speech signal even further. Noise PSD estimation algorithms using a priori information about the noise statistics have been shown to track non-stationary noise better than the conventional algorithms which rely on the minimum statistics approach.In this thesis, we perform noise suppression in the modulation domain with the noise and speech PSD derived from an estimation scheme which employs the a priori information of various speech and noise types. Specifically, codebooks of gain normalized linear prediction coefficients obtained from training on various speech and noise files are used as the a priori information while performing the estimation of the desired PSD. The PSD estimates derived from this codebook approach are used to obtain a minimum mean square error (MMSE) estimate of the clean speech modulation magnitude spectrum, which is then combined with the phase spectrum of the noisy speech to recover the enhanced speech signal. The enhanced speech signal is subjected to various objective experiments for evaluation. Results of these evaluations indicate improvement in noise suppression with the proposed codebook-based modulation domain approach over competing approaches, particularly in cases of nonstationary noise.ii Sommaire Les méthodes conventionnelles de rehaussement de la paroleà canal unique utilisent une structure d'analyse-modification-synthèse (AMS) dans le domaine fréquentiel. Récemment, il aété démontré que l'utilisation de cette structure dans le domaine de la modulation pourrait offrir une meilleure suppression du bruit. Toutefois, cette conclusion aété obtenue en se basant sur une approcheà statistique minimale pour l'estimation de la densité spectrale de puissance (PSD) du bruit ambiant.Plusieurs algorithmes d'estimation de bruit ontété proposés au fil des ans dans la littérature sur le traitement de la parole. D'ordinaire, les méthodes d'estimation de bruit qui se servent d'une approcheà statistique minimale vont créer un décalage temporel dans l'estimation spectrale du bruit. Ce décalage peut engendrer beaucou...

show abstract

“…. , y T , S T = H k in Equation 7. In this section, we are interested in exploiting memory to ensure that the codebook that is most relevant to the current context at hand receives a high likelihood, and this is captured by Equation 11.…”

Section: Bayesian Estimation Under Varying Contextsmentioning

confidence: 99%

“…Herein, an estimate of the noise magnitude spectrum is subtracted from the observed noisy magnitude spectrum to obtain an estimate of the clean speech magnitude spectrum. Several variations of this technique have been developed over the years [4][5][6][7]. Methods based on a statistical model of speech to estimate the speech spectral amplitude such as the minimum mean square error shorttime spectral amplitude estimator (MMSE-STSA) method have been found to be successful [8][9][10].…”

Section: Introductionmentioning

confidence: 99%

Robust Bayesian estimation for context-based speech enhancement

Naidu

Srinivasan

2014

J AUDIO SPEECH MUSIC PROC.

View full text Add to dashboard Cite

Model-based speech enhancement algorithms that employ trained models, such as codebooks, hidden Markov models, Gaussian mixture models, etc., containing representations of speech such as linear predictive coefficients, mel-frequency cepstrum coefficients, etc., have been found to be successful in enhancing noisy speech corrupted by nonstationary noise. However, these models are typically trained on speech data from multiple speakers under controlled acoustic conditions. In this paper, we introduce the notion of context-dependent models that are trained on speech data with one or more aspects of context, such as speaker, acoustic environment, speaking style, etc. In scenarios where the modeled and observed contexts match, context-dependent models can be expected to result in better performance, whereas context-independent models are preferred otherwise. In this paper, we present a Bayesian framework that automatically provides the benefits of both models under varying contexts. As several aspects of the context remain constant over an extended period during usage, a memory-based approach that exploits information from past data is employed. We use a codebook-based speech enhancement technique that employs trained models of speech and noise linear predictive coefficients as an example model-based approach. Using speaker, acoustic environment, and speaking style as aspects of context, we demonstrate the robustness of the proposed framework for different context scenarios, input signal-to-noise ratios, and number of contexts modeled.

show abstract

Speech enhancement using a minimum mean-square error short-time spectral modulation magnitude estimator

Cited by 64 publications

References 35 publications

Spectral and Cepstral Audio Noise Reduction Techniques in Speech Emotion Recognition

Spectral and Cepstral Audio Noise Reduction Techniques in Speech Emotion Recognition

Speech enhancement in modulation domain using codebook-based speech and noise estimation

Robust Bayesian estimation for context-based speech enhancement

Contact Info

Product

Resources

About