Improving a GMM speaker verification system by phonetic weighting

Auckenthaler, Roland; Parris, Eluned S.; Carey, Michael J.

doi:10.1109/icassp.1999.758125

Cited by 43 publications

(29 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, we need to fix the length of the test segment to be equal for all classes for fair comparison. Previous studies have fixed the length of the test segment [1], [3] or showed the results as an average likelihood versus the number of frames in each phoneme [2] for a similar reason. Thus, we also perform speaker identification test with a fixed test length.…”

Section: Controlled Condition: Same Test Lengthmentioning

confidence: 99%

“…Fundamentally, they use the fact that some classes of phonemes, such as vowels and nasals, include more speaker-related information than obstruents such as stops, affricatives, and fricatives [1]. One of the approaches assigns different weights to each phoneme depending on how much each phoneme includes speaker related information [2]. Pelecanos et al found an optimal feature type and a frame length of each phoneme for speaker identification [3].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

On the Importance of Transition Regions for Automatic Speaker Recognition

Lee

Jung

Choi

et al. 2010

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYThis letter describes the importance of transition regions, e.g. at phoneme boundaries, for automatic speaker recognition compared with using steady-state regions. Experimental results of automatic speaker identification tasks confirm that transition regions include the most speaker distinctive features. A possible reason for obtaining such results is described in view of articulation, in particular, the degree of freedom of articulators. These results are expected to provide useful information in designing an efficient automatic speaker recognition system.

show abstract

Section: Controlled Condition: Same Test Lengthmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

On the Importance of Transition Regions for Automatic Speaker Recognition

Lee

Jung

Choi

et al. 2010

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

show abstract

“…[4][5][6][7] In this paper, we also propose a method that utilizes phoneme information to improve speaker recognition performance. While previous studies focus on using separate models for each phoneme and combining scores, 4,6,7 we focus on finding an optimal phoneme class ratio, the portion of each phoneme class in an utterance, that maximizes speaker recognition performance based on mutual information. In speaker recognition, some researchers use this measurement to measure or improve speaker recognition accuracy.…”

Section: Introductionmentioning

confidence: 99%

Phonetically optimized speaker modeling for robust speaker recognition

Lee

Choi

Kang

2009

The Journal of the Acoustical Society of America

View full text Add to dashboard Cite

This paper proposes an efficient method to improve speaker recognition performance by dynamically controlling the ratio of phoneme class information. It utilizes the fact that each phoneme contains different amounts of speaker discriminative information that can be measured by mutual information. After classifying phonemes into five classes, the optimal ratio of each class in both training and testing processes is adjusted using a non-linear optimization technique, i.e., the Nelder–Mead method. Speaker identification results verify that the proposed method achieves 18% improvement in terms of error rate compared to a baseline system.

show abstract

“…The Universal Background Model (UBM) approach [9] is the most popular normalized likelihood approach when utilizing Gaussian Mixture Model (GMM) classifier.…”

Section: Likelihood Normalizationmentioning

confidence: 99%

“…The background models were then adapted to speaker models using MAP adaptation [9]. The first two utterances for all speakers in the corpus being common were used for text dependent experiments and 6 different utterances for each speaker allowed text independent verification experiments to be conducted.…”

Section: Likelihood Normalizationmentioning

confidence: 99%

Investigating feature-level fusion for checking liveness in face-voice authentication

Chetty¹,

Wagner²

Proceedings of the Eighth International Symposium on Signal Processing and Its Applications, 2005.

View full text Add to dashboard Cite

In this paper we propose a feature level fusion approach for checking liveness in face-voice person authentication.Liveness verification experiments conducted on two audiovisual databases, VidTIMIT and UCBN, show that feature-level fusion is indeed a powerful technique for checking liveness in systems that are vulnerable to replay attacks, as it preserves synchronisation between closely coupled modalities, such as voice and face, through various stages of authentication. An improvement in error rate of the order of 25-40% is achieved for replay attack experiments by using feature level fusion of acoustic and visual feature vectors from lip region as compared to classical late fusion approach.

show abstract

Improving a GMM speaker verification system by phonetic weighting

Cited by 43 publications

References 3 publications

On the Importance of Transition Regions for Automatic Speaker Recognition

On the Importance of Transition Regions for Automatic Speaker Recognition

Phonetically optimized speaker modeling for robust speaker recognition

Investigating feature-level fusion for checking liveness in face-voice authentication

Contact Info

Product

Resources

About