Velocity and acceleration features in speaker recognition

Mason, John S.; Zhang, X.

doi:10.1109/icassp.1991.151073

Cited by 28 publications

(8 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The second ones (denoted with ∆∆ f , as the vector of ∆∆ coefficients for the f -th frame) are obtained by performing again the time derivatives on the first ∆ f coefficients. Usually, these coefficients are termed as velocityand acceleration-coefficients, respectively [6]. The motivation to employ, for each frame f , only the MFCC and ∆∆ coefficients relies on the fact that the feature vector X f = [C f ∆∆ f ] (obtained by concatenating the MFCC vector C f and the ∆∆ f one) provides the best results in terms of trade-off between percentage of correct recognitions and time needed to perform a single recognition.…”

Section: A Front-end and Feature Extractionmentioning

confidence: 99%

“…The problem of recognizing a speaker among a close-set of speakers is widely studied in the speech processing literature (see, among the others [3], [6] and references therein). The main difference between this work and most of the other works in the literature is that the speaker recognition algorithm embedded within SPECTRA is able to dynamically increase the number of recognizable speakers.…”

Section: B Speaker Recognitionmentioning

confidence: 99%

See 1 more Smart Citation

SPECTRA: A SPEech proCessing plaTform as smaRtphone Application

Bisio

Lavagetto

Marchese

et al. 2015

2015 IEEE International Conference on Communications (ICC)

View full text Add to dashboard Cite

In this paper, an Android SPEech proCessing plaTform as smaRtphone Application (SPECTRA) is presented. Such application, developed by the authors, has three main functions: i) Gender Recognition (GR), ii) Speaker Recognition (SR) and iii) Language Recognition (LR). All these recognition functions are performed simultaneously by using unsupervised Support Vector Machine (SVM) classifiers. An innovative point of this paper lies in the automatic re-training of the employed SVMs which are able to dynamically update themselves when a (new) audio from a (new) speaker is provided. This allow to build more robust classifiers, which results in better recognition performances. In terms of accuracy, the GR reaches about 98% of correct classifications, SR performs around 80% while LR shows an accuracy of about 74%.

show abstract

Section: A Front-end and Feature Extractionmentioning

confidence: 99%

Section: B Speaker Recognitionmentioning

confidence: 99%

SPECTRA: A SPEech proCessing plaTform as smaRtphone Application

Bisio

Lavagetto

Marchese

et al. 2015

2015 IEEE International Conference on Communications (ICC)

View full text Add to dashboard Cite

show abstract

“…A widely method to encode some of the dynamic information over time of spectral features is known as delta features "∆" [3,5]. The time derivatives of each cepstral coefficient are obtained by differentiation and zero padding at begin and end of the utterance, then, the estimate of the derivative is appended to the acoustic vector, yielding a higher-dimensional feature vector.…”

Section: Extracting Delta and Delta-delta Featuresmentioning

confidence: 99%

A Method of Automatic Speaker Recognition Using Cepstral Features and Vectorial Quantization

Lara

2005

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Automatic Speaker Recognition techniques are increasing the use of the speaker's voice to control access to personalized telephonic services. This paper describes the use of vector quantization as a feature matching method, in an automatic speaker recognition system, evaluated with speech samples from a SALA Spanish Venezuelan database for fixed telephone network. Results obtained reflect a good performance of the method in a text independent job in the context of sequences of digits.

show abstract

“…Due to the importance of such parameters, many researchers focused on improving them and trying to capture dynamic information more efficiently. One example is the use of regression features [11,12] instead of first and second order derivatives. Another recently proposed technique that uses DCT-based contextualization [13] proposes to replace MFCC features and their derivatives by a 2D-DCT transform applied on the Mel filter bank outputs.…”

Section: Introductionmentioning

confidence: 99%

Local binary patterns as features for speaker recognition

Kheder¹,

Matrouf²,

Ajili³

et al. 2016

The Speaker and Language Recognition Workshop (Odyssey 2016)

View full text Add to dashboard Cite

The i-vector framework witnessed great success in the past years in speaker recognition (SR). The feature extraction process is central in SR systems and many features have been developed over the years to improve the recognition performance. In this paper, we present a new feature representation which borrows a concept initially developed in computer vision to characterize textures called Local Binary Patterns (LBP). We explore the use of LBP as features for speaker recognition and show that using them as descriptors for cepstral coefficients dynamics (replacing ∆ and ∆∆ in the regular MFCC representation) results in more efficient features and yield up to 15% of relative improvement compared to the baseline system performance in both clean and noisy conditions.

show abstract

Velocity and acceleration features in speaker recognition

Cited by 28 publications

References 2 publications

SPECTRA: A SPEech proCessing plaTform as smaRtphone Application

SPECTRA: A SPEech proCessing plaTform as smaRtphone Application

A Method of Automatic Speaker Recognition Using Cepstral Features and Vectorial Quantization

Local binary patterns as features for speaker recognition

Contact Info

Product

Resources

About