An algorithm that improves speech intelligibility in noise for normal-hearing listeners.

By Gibak Kim, Yang Lu, Y. Hu, P. Loizou. Journal of the Acoustical Society of America, 2009

365 citations recorded by the source corpus.

Subfields: Cognitive Neuroscience, Computational Mechanics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/52bc2f301aa791d2391882971b9969d563b8e2c4

How researchers see this paper

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

In Cognitive Neuroscience

In Computational Mechanics

In Signal Processing

Citations in the corpus

Complex Ratio Masking for Monaural Speech Separation (2015, background): “Another factor that motivates us to examine phase estimation is that supervised mask estimation has recently been shown to improve human speech intelligibility in very noisy conditions [8], [9].”
On Training Targets for Supervised Speech Separation (2014, background): “In many previous studies (e.g. [10], [16], [17]), the training target (or the learning signal) is set to the ideal binary mask (IBM), which is a binary mask constructed from premixed speech and noise signals (see Section III-A for de ﬁ nition).”
Investigation of Speech Separation as a Front-End for Noise Robust Speech Recognition (2014, methodology): “Stereo training data is also used by supervised classi ﬁ cation based speech separation algorithms [33], [36], [22], [17], [43].”
Exploring Monaural Features for Classification-Based Speech Segregation (2013, methodology): “A positive effect of adding delta features with AMS has been shown in [23].”
Reasons why current speech-enhancement algorithms do not improve speech intelligibility and suggested solutions (2011, result): “This was confirmed in two listening studies with IdBM-processed speech [28], [29] and in one study with estimated time–frequency masks [34].”
Deep clustering: Discriminative embeddings for segmentation and separation (2015, background): “In the audio domain, single-channel speech separation methods, for example, segment the time-frequency elements of the spectrogram into regions dominated by a target speaker, either based on classifiers [6–8], or generative models [9–11].”
Cocktail Party Processing via Structured Prediction (2012, methodology): “Speech intelligibility studies [9] [10] have evaluated the influence of the hit (HIT) and false-alarm (FA) rate on intelligibility scores.”
Binaural Classification for Reverberant Speech Segregation Using Deep Neural Networks (2014, methodology): “The HIT FA rate has been shown to be correlated to human intelligibility [20].”
A deep neural network for time-domain signal reconstruction (2015, methodology): “The joint training of speech resynthesis and mask estimation yields improved objective quality while maintaining the objective intelligibility performance.”
Deep Belief Networks Based Voice Activity Detection (2013, methodology): “We only use 3 bands of the AMS features with two delta features [48], which is much smaller than the AMS features used in [48], this is mainly for the detection efficiency.”
Towards Scaling Up Classification-Based Speech Separation (2013, background): “Speech intelligibility studies [30], [34] have shown that falsealarm (FA) errors are far more detrimental to human speech intelligibility than miss errors.”
Learning Spectral Mapping for Speech Dereverberation and Denoising (2015, methodology): “The training set contains reverberant mixtures including 200 utterances convolved with 3 RIRs and mixed with 3 noise types: speech-shaped noise, factory noise and babble noise [16] at 0 dB SNR.”
Speech intelligibility in reverberation with ideal binary masking: effects of early reflections and signal-to-noise ratio threshold. (2013, methodology): “…to estimate the IBM from the mixture, the binary masking approach allows the use of supervised learning algorithms to perform classiﬁ-cation, which are proving to be powerful alternatives to more traditional signal enhancement methods (Roman et al. , 2003; Kim et al. , 2009; Han and Wang, 2011).”
An algorithm to improve speech recognition in noise for hearing-impaired listeners. (2013, methodology): “Previous classification-based segregation systems have employed GMM (Kim et al., 2009) or support vector machines (SVM, Han and Wang, 2012).”
A Deep Ensemble Learning Method for Monaural Speech Separation (2016, methodology): “Kim et al. [20] used Gaussian mixture models (GMM) to learn the distribution of target and interference dominant T-F units and then built a Bayesian classiﬁer to estimate the IBM.”