i-Vector/HMM Based Text-Dependent Speaker Verification System for RedDots Challenge

Zeinali, Hossein; Sameti, Hossein; Burget, Lukáš; Černocký, Jaň; Maghsoodi, Nooshin; Matějka, Pavel

doi:10.21437/interspeech.2016-1174

Cited by 27 publications

(27 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 1 compares performance of relevance MAP, i-vector and JFA systems exploiting posteriors estimated using GMM-UBM. It can be seen that RMAP GMM achieves the best results among the modelbased SV systems, which is consistent with the results in [20]. Although in [14], the JFA GMM outperformed RMAP GMM and Ivec Table 3 shows the performance of the DTW SV systems (DTW-MFCC, DTW-post GMM , DTW-post DNN , DTW-onIvec GMM and DTW-onIvec DNN ) using different features.…”

Section: Resultssupporting

confidence: 79%

See 1 more Smart Citation

Exploiting sequence information for text-dependent Speaker Verification

Dey

Motlíček

Madikeri

et al. 2017

2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Model-based approaches to Speaker Verification (SV), such as Joint Factor Analysis (JFA), i-vector and relevance Maximum-aPosteriori (MAP), have shown to provide state-of-the-art performance for text-dependent systems with fixed phrases. The performance of i-vector and JFA models has been further enhanced by estimating posteriors from Deep Neural Network (DNN) instead of Gaussian Mixture Model (GMM). While both DNNs and GMMs aim at incorporating phonetic information of the phrase with these posteriors, model-based SV approaches ignore the sequence information of the phonetic units of the phrase. In this paper, we tackle this issue by applying dynamic time warping using speaker-informative features. We propose to use i-vectors computed from short segments of each speech utterance, also called online i-vectors, as feature vectors. The proposed approach is evaluated on the RedDots database and provides an improvement of 75% relative equal error rate over the best model-based SV baseline system in a content-mismatch condition.

show abstract

Section: Resultssupporting

confidence: 79%

“…Since the relevance MAP approach has shown to provide good results on RedDots data [20], we consider this system as the baseline. First, we analyse the performance of the model-based SV approaches and then the DTW systems.…”

Section: Resultsmentioning

confidence: 99%

Exploiting sequence information for text-dependent Speaker Verification

Dey

Motlíček

Madikeri

et al. 2017

2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

show abstract

“…Table 8 compares the performance of all systems on RedDots dataset across all the conditions. We consider the MAP system (MAP GMM ) using GMM posterior as the baseline since it has shown to provide good performance in Zeinali et al (2016). The model-based SV systems perform worse on the RedDots database compared to RSR database (Dey et al, 2016a).…”

Section: Summary Of Experiments On Rsr Databasementioning

confidence: 99%

Template-matching for text-dependent speaker verification

Dey

Motlíček

Madikeri

et al. 2017

Speech Communication

View full text Add to dashboard Cite

show abstract

“…These systems consider the feature dynamics of the words for identification. The most common modeling techniques for text-dependent speaker recognition are the Hidden Markov Models (HMM) [26] and Dynamic Time Warping (DTW) [27].…”

Section: State Of the Artmentioning

confidence: 99%

“…MFCC is a filterbankbased approach designed to resemble the human auditory frequency perception. Other feature extraction methods are: delta-MFCC and delta-delta MFCC [32], linear predictive cepstral coefficients [33], perceptual linear prediction [34], coefficients cepstral mean and variance normalization [35], relative spectral transform filtering [36], feature warping [27], i-vectors and super-vectors [26].…”

Section: State Of the Artmentioning

confidence: 99%

Text-independent speaker recognition for Ambient Intelligence applications by using Information Set Features

Anand

Labati

Hanmandlu

et al. 2017

2017 IEEE International Conference on Computational Intelligence and Virtual Environments for Measurement Systems and Applicati

View full text Add to dashboard Cite

Abstract-Biometric systems are enabling technologies for a wide set of applications in Ambient Intelligence (AmI) environments. In this context, speaker recognition techniques are of paramount importance due to their high user acceptance and low required cooperation. Typical applications of biometric recognition in AmI environments are identification techniques designed to recognize individuals in small datasets. Biometric recognition methods are frequently deployed on embedded hardware and therefore need to be optimized in terms of computational time as well as used memory. This paper presents a textindependent speaker recognition method particularly suitable for identification in AmI environments. The proposed method first computes the Mel Frequency Cepstral Coefficients (MFCC) and then creates Information Set Features (ISF) by applying a fuzzy logic approach. Finally, it estimates the user's identity by using a hierarchical classification technique based on computational intelligence. We evaluated the performance of the speaker recognition method using signals belonging to the NIST-2003 switchboard speaker database. The achieved results showed that the proposed method reduced the size of the template with respect to traditional approaches based on Gaussian Mixture Models (GMM) and achieved better identification accuracy.

show abstract

i-Vector/HMM Based Text-Dependent Speaker Verification System for RedDots Challenge

Cited by 27 publications

References 6 publications

Exploiting sequence information for text-dependent Speaker Verification

Exploiting sequence information for text-dependent Speaker Verification

Template-matching for text-dependent speaker verification

Text-independent speaker recognition for Ambient Intelligence applications by using Information Set Features

Contact Info

Product

Resources

About