Using blind source separation techniques to improve speech recognition in bilateral cochlear implant patients

Kokkinakis, Kostas; Loizou, Philipos C.

doi:10.1121/1.2839887

Cited by 37 publications

(20 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Approaches that address the problem of estimating the underlying source signals of a mixture [1] can be extremely useful for speech recognition systems [2], in medical signal processing [3], [4]; as well as in hearing aid devices [5], [6]. The problem becomes more challenging when only monaural mixture recordings are available [7] and therefore is still an active area of research.…”

Section: Introductionmentioning

confidence: 99%

Monaural Source Separation Using a Random Forest Classifier

et al. 2016

View full text Add to dashboard Cite

We address the problem of separating two audio sources from a single channel mixture recording. A novel method called Multi Layered Random Forest (MLRF) that learns a binary mask for both the sources is presented. Random Forest (RF) classifiers are trained for each frequency band of a source spectrogram. A specialized set of linear transformations are applied to a local time-frequency (T-F) neighborhood of the mixture that captures relevant local statistics. A sampling method is presented that efficiently samples T-F training bins in each frequency band. We draw equal numbers of dominant (more power) training samples from the two sources for RF classifiers that estimate the Ideal Binary Mask (IBM). An estimated IBM in a given layer is used to train a RF classifier in the next higher layer of the MLRF hierarchy. On average, MLRF performs better than deep Recurrent Neural Networks (RNNs) and Non-Negative Sparse Coding (NNSC) in signal-to-noise ratio (SNR) of reconstructed audio, overall T-F bin classification accuracy, as well as PESQ and STOI scores. Additionally, we demonstrate the ability of the MLRF to correctly reconstruct T-F bins of the target even when the latter has lower power in that frequency band. ABSTRACTWe address the problem of separating two audio sources from a single channel mixture recording. A novel method called Multi Layered Random Forest (MLRF) that learns a binary mask for both the sources is presented. Random Forest (RF) classifiers are trained for each frequency band of a source spectrogram. A specialized set of linear transformations are applied to a local time-frequency (T-F) neighborhood of the mixture that captures relevant local statistics. A sampling method is presented that efficiently samples T-F training bins in each frequency band. We draw equal numbers of dominant (more power) training samples from the two sources for RF classifiers that estimate the Ideal Binary Mask (IBM). An estimated IBM in a given layer is used to train a RF classifier in the next higher layer of the MLRF hierarchy. On average, MLRF performs better than deep Recurrent Neural Networks (RNNs) and Non-Negative Sparse Coding (NNSC) in signalto-noise ratio (SNR) of reconstructed audio, overall T-F bin classification accuracy, as well as PESQ and STOI scores. Additionally, we demonstrate the ability of the MLRF to correctly reconstruct T-F bins of the target even when the latter has lower power in that frequency band.

show abstract

Section: Introductionmentioning

confidence: 99%

Monaural Source Separation Using a Random Forest Classifier

et al. 2016

View full text Add to dashboard Cite

show abstract

“…While being an interesting problem in itself, the separation of sources from a mixture can serve as a intermediary step for other tasks such as automatic speech recognition, [9] and fundamental frequency estimation, [5]. Some applications, such as speech enhancement for cochlear implant users, [9] [7], require low-latency processing, which we will focus on in this paper.…”

Section: Introductionmentioning

confidence: 99%

Monoaural Audio Source Separation Using Deep Convolutional Neural Networks

Chandna

Miron

Janer

et al. 2017

Lecture Notes in Computer Science

160

123

View full text Add to dashboard Cite

Abstract. In this paper we introduce a low-latency monaural source separation framework using a Convolutional Neural Network (CNN). We use a CNN to estimate time-frequency soft masks which are applied for source separation. We evaluate the performance of the neural network on a database comprising of musical mixtures of three instruments: voice, drums, bass as well as other instruments which vary from song to song. The proposed architecture is compared to a Multilayer Perceptron (MLP), achieving on-par results and a significant improvement in processing time. The algorithm was submitted to source separation evaluation campaigns to test efficiency, and achieved competitive results.

show abstract

“…Specifically, in a sound-proof room with straightahead target speech and noise at 90 , 3 of 11 listeners showed worse performance with the processing, while the remaining 8 had speech-reception improvements ranging from 20 to 77 percentage points. More recently, binaural noise reduction has been shown to improve performance for CI users (Goldsworthy, 2005;Kokkinakis and Loizou, 2008) with speech reception benefits in noise as large as 60% on keyword recognition and benefits of 14 dB in SRTs. Goldsworthy et al (2009) transitioned this style of spatial filtering to a configuration with two closely-spaced ($2 cm) microphones mounted in a behind-the-ear (BTE) shell; they argued that this approach was more robust to the effects of reverberation and demonstrated speech reception benefits for CI users.…”

Section: Introductionmentioning

confidence: 99%

Two-microphone spatial filtering provides speech reception benefits for cochlear implant users in difficult acoustic environments

Goldsworthy

Delhorne

Desloge

et al. 2014

The Journal of the Acoustical Society of America

View full text Add to dashboard Cite

This article introduces and provides an assessment of a spatial-filtering algorithm based on two closely-spaced ($1 cm) microphones in a behind-the-ear shell. The evaluated spatial-filtering algorithm used fast ($10 ms) temporal-spectral analysis to determine the location of incoming sounds and to enhance sounds arriving from straight ahead of the listener. Speech reception thresholds (SRTs) were measured for eight cochlear implant (CI) users using consonant and vowel materials under three processing conditions: An omni-directional response, a dipole-directional response, and the spatial-filtering algorithm. The background noise condition used three simultaneous time-reversed speech signals as interferers located at 90 , 180 , and 270. Results indicated that the spatial-filtering algorithm can provide speech reception benefits of 5.8 to 10.7 dB SRT compared to an omni-directional response in a reverberant room with multiple noise sources. Given the observed SRT benefits, coupled with an efficient design, the proposed algorithm is promising as a CI noise-reduction solution.

show abstract

Using blind source separation techniques to improve speech recognition in bilateral cochlear implant patients

Cited by 37 publications

References 48 publications

Monaural Source Separation Using a Random Forest Classifier

Monaural Source Separation Using a Random Forest Classifier

Monoaural Audio Source Separation Using Deep Convolutional Neural Networks

Two-microphone spatial filtering provides speech reception benefits for cochlear implant users in difficult acoustic environments

Contact Info

Product

Resources

About