Improving speech intelligibility in cochlear implants using vocoder-centric acoustic models

Gladston, Anushiya Rachel; Vijayalakshmi, P.; Nagarajan, T.

doi:10.1109/icrtit.2012.6206795

Cited by 3 publications

(4 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For this reason, we proposed to deduce this JD denoted as JD (1,2) N AR,k from the one computed when the processes are noise-free. More particularly, using the inversion matrix lemma [23] we showed that:…”

Section: Definition Of Jeffrey's Divergence Between Stochastic Pmentioning

confidence: 99%

“…The JD between the process ARMA(a (1) , b (1) , σ 2 (1) ) and the process ARMA(a (2) , b (2) , σ 2 (2) ) amounts to studying the JD between the process NAR(a (1) , , Ψ (1) , Υ (1) ) and the process NAR(a (2) , , Ψ (2) , Υ (2) ) where Ψ (l) and Υ (l) with l = 1, 2 are defined as in (4). In the following, our purpose is to express the increment as we did in (26):…”

Section: Jd Between Arma Processes Based On the Analysis Of The Jmentioning

confidence: 99%

“…Another way is to compare the power spectra of the data. Thus, the spectral distortion between the natural and the synthetic speech can be based on the COSH distance which is defined from the power spectra of the data to be compared [2]. The log-spectral distance and the Itakura-Saito distance are also widely computed, especially in speech processing.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Jeffrey's divergence between autoregressive moving-average processes

Legrand¹,

Grivel²,

Giremus³

2017

2017 25th European Signal Processing Conference (EUSIPCO)

View full text Add to dashboard Cite

Abstract-Various works have been carried out about the Jeffrey's divergence (JD) which is the symmetric version of the Kullback-Leibler (KL) divergence. An expression of the JD for Gaussian processes can be deduced from the definition of the KL divergence and the expression of the Gaussian-multivariate distributions of k-dimensional random vectors. It depends on the k × k Toeplitz covariance matrices of the stationary processes. However, the resulting computational cost may be high as these matrices must be inverted and it is all the higher as k increases. To circumvent this problem, a recursive expression can be obtained for real 1 st -order autoregressive (AR) processes. When they are disturbed by additive uncorrelated white noises, we showed that when k becomes large, the derivative of the JD with respect to k tends to be constant. This constant is sufficient to compare the noisy AR processes. In this paper, we propose to extend our work to AR moving-average (MA) processes with one AR term and one MA term. Some examples illustrate the theoretical analysis.

show abstract

Section: Definition Of Jeffrey's Divergence Between Stochastic Pmentioning

confidence: 99%

Section: Jd Between Arma Processes Based On the Analysis Of The Jmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Jeffrey's divergence between autoregressive moving-average processes

Legrand¹,

Grivel²,

Giremus³

2017

2017 25th European Signal Processing Conference (EUSIPCO)

View full text Add to dashboard Cite

show abstract

“…hearing improvement= f target f total (12) In this case, an example is bringing frequencies 3600 -4000 in a maximum 4000 frequency sample into the audible range by high compression. The improvement would be 400/4000, or 10%, meaning that 10% of the sample has now been changed and brought into the audible range.…”

Section: Hearing Metricsmentioning

confidence: 99%

Intelligent Hearing Assistance Using Self-Organizing Feature Maps

Hodges

Zohdy

2014

TMLAI

View full text Add to dashboard Cite

A novel hearing assistance system is proposed which classifies sounds and selectively tunes them according to the needs of the hearing impaired. This differs from the usual hearing aids available today in that it uses computational intelligence to filter and tune sounds based on real life categorical classifications. The system can significantly improve audibility for the hearing impaired, including bringing completely inaudible tones into audibility. For classification a self-organizing feature map is used with a vector of sound features from the joint time-frequency domain. The map is trained with input sounds until a map of neurons is clustered according to these. The resulting map is used to classify new sounds. Based on this classification, a sound can be tuned to improve audibility for the hearing impaired. Techniques proposed for audio output include gamma tone frequency filtering, Fourier compression, low pass filtering, spectral subtraction, and an original algorithm to choose how to best boost the amplitude of deaf frequencies.

show abstract

LabVIEW and digital signal processor implementation of a channel vocoder based model of a cochlear implant

Rachel¹,

Singh²,

Vijayalakshmi³

2013

2013 International Conference on Recent Trends in Information Technology (ICRTIT)

Self Cite

View full text Add to dashboard Cite

A cochlear implant is a prosthetic device used to mimic the function of a cochlea in a person with profound and bilateral hearing loss caused by a damaged inner ear. The current work revolves around the design of real time channel vocoder based model of a cochlear implant in LabVIEW and the TMS320C6713 DSK. First, a uniform band 16-channel vocoder is designed for the analysis and synthesis of English vowels, where filters of 400 Hz bandwidth, with cut off frequencies up to 6200Hz are used, based on MATLAB analysis performed previously. To extend the analysis to words and sentences, short time features, namely, short time energy, short time zero crossing rate and main lobe width of the short time autocorrelation function, are extracted and Gaussian Mixture Modelling (GMM) is used to classify the speech segments as voiced or unvoiced. In an attempt to make the synthetic speech sound natural, the synthesis in the channel vocoder is done using a train of glottal pulses instead of a train of impulses. The intelligibility of the synthetic speech is measured by the Mean Opinion Score (MOS). For a channel vocoder where the synthesis section uses a train of glottal pulses, an MOS of 3.6 is obtained as against 3.5 when a train of impulses is used. The lab model of the cochlear implant, that is, the analysis section of the 16-channel vocoder is then realised in the TMS320C6713 DSK. Individual channel outputs are obtained by programming the DIP switches of the DSK and a DSK_AUDIO16_BASE, a 16-channel audio daughter card, is interfaced with the DSK to obtain the outputs from multiple channels simultaneously.

show abstract

Improving speech intelligibility in cochlear implants using vocoder-centric acoustic models

Cited by 3 publications

References 14 publications

Jeffrey's divergence between autoregressive moving-average processes

Jeffrey's divergence between autoregressive moving-average processes

Intelligent Hearing Assistance Using Self-Organizing Feature Maps

LabVIEW and digital signal processor implementation of a channel vocoder based model of a cochlear implant

Contact Info

Product

Resources

About