Enhancement of speech corrupted by acoustic noise

By M. Berouti, Richard M. Schwartz, John I. Makhoul. 1979

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Y. Hu
2. P. Loizou
3. Santiago Pascual
4. Kuldip K. Paliwal
5. Joan Serrà
6. Chandan K. A. Reddy
7. Deliang Wang
8. Israel Cohen

Citations in the corpus

• SEANet: A Multi-modal Speech Enhancement Network (2020, neutral): “Traditional signal processing methods adopt spectral noise subtraction [1, 2], spectral masking [3, 4], statistical methods based on Wiener filtering [5] and Bayesian estimators [6, 7].”
• SEGAN: Speech Enhancement Generative Adversarial Network (2017, methodology): “Classic speech enhancement methods are spectral subtraction [6], Wiener filtering [7], statistical model-based meth-ods [8], and subspace algorithms [9, 10].”
• An individualized super-Gaussian single microphone Speech Enhancement for hearing aid users with smartphone as an assistive device (2017, methodology): “Traditional methods such as spectral subtraction [6] introduce musical noise due to the half-wave rectification problem [7], which is prominent at lower SNRs.”
• Speech enhancement for non-stationary noise environments (2001, background): “…by ˆ A(k; ') = (exp{E[log A(k; ')|Y (k; '); H 1 (k; ')]}) p(k; ') × (exp{E[log A(k; ')|Y (k; '); H 0 (k; ')]}) (1−p(k; ')) : (12) When speech is absent, the gain is constrained to be larger than a threshold G min , which is determined by a subjective criteria for the noise naturalness [1] [2] [29].”
• Efficient voice activity detection algorithms using long-term speech information (2004, methodology): “There exist well known noise suppression algorithms (Berouti et al., 1979; Boll, 1979), such as Wiener filtering (WF) or spectral subtraction, that are widely used for robust speech recognition, and for which, the VAD is critical in attaining a high level of performance .”
• Noise power spectral density estimation based on optimal smoothing and minimum statistics (2001, unknown): “Even though most speech enhancement algorithms use a modified noise psd (noise " overestimation " [18] or noise " underestimation " [19]) we believe it is of utmost importance to first obtain an unbiased noise psd estimate and then to modify it based on statistical arguments or on listening tests.”
• A multi-band spectral subtraction method for enhancing speech corrupted by colored noise (2002, methodology): “Finally, the standard overlap-and-add method is used to obtain the enhanced signal.”
• Spectral Subtraction Based on Minimum Statistics (2001, unknown): “[4] we subtract spectral magnitudes with an oversubtraction factor “?”‡•’–?z {#|*}o~ and a limitation of the maximum subtraction by a spectral floor constant ”3•6–Z— ( „ … „o[˜™”!•6–‡—^˜š„ … „ › )  œz {#|*}+~‡ ˆ ”!•6–‡—žŠ x y z {#|*}+~ Ÿ —  €]z { |¡} ~Z Š3¢žz {#|¡} ~‘˜ ”3•6–‡—žŠ x y z {#|*}o~ ”
• Speech enhancement with missing data techniques using recurrent neural networks (2004, methodology): “The enhancement system consists of two stages: the first stage uses a Recurrent Neural Network, which is supplied with noisy speech and produces enhanced speech; whereas the second stage uses missing data techniques to further improve the quality of enhanced speech.”
• A New Method Based on Spectral Subtraction for Speech Dereverberation (2001, methodology): “[41, 38]) in order to suppress more noise.”
• Subjective comparison and evaluation of speech enhancement algorithms (2007, methodology): “…< 8 1:5; i ¼ 8 8><>: ð2Þ The band-specific subtraction factor ai is a piecewise linear function of the segmental SNR of band i and is calculated as follows (Berouti et al., 1979): ai ¼ 5; SNRi < 5 4 3 20 ðSNRiÞ; 5 6 SNRi 6 20 1; SNRi > 20 8><>: ð3Þ where SNRi ðdBÞ ¼ 10log10 Pei k¼bi jbY iðkÞj2Pei…”
• Overview of speech enhancement techniques for automatic speaker recognition (1996, background): “In the NSS technique, an estimate of both noise and speech can be derived from the following expresions, R R R i R i R i ( ) ( ) ( ) ( ) ω λ ω λ ω = + − −1 1 (4)”
• An open access database for the evaluation of respiratory sound classification algorithms (2019, methodology): “The stationary noise in audio files was suppressed using spectral subtraction (SS) (Berouti et al 1979).”
• The importance of phase in speech enhancement (2011, methodology): “These can be grouped into spectral subtraction (Boll, 1979; Berouti et al., 1979), minimum mean-square error (MMSE) estimation (Ephraim and Malah, 1984, 1985), Wiener filtering (linear MMSE) (Wiener, 1949), Kalman filtering (Paliwal and Basu, 1987) and subspace (Ephraim and Trees, 1995) methods.”