Probabilistic Latent Prosody Analysis for Robust Speaker Verification

Chen, Zi-He; Zeng, Zhi-Ren; Liao, Yuan-Fu; Juang, Yau-Tarng

doi:10.1109/icassp.2006.1659968

Cited by 3 publications

(5 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To address this problem, higher level information such as the prosodic cues of a speaker, which may be less sensitive to those mismatch, are attractive recently. For example, several works [1][2][3] have shown there is a significant benefit to combining prosodic and spectral features.…”

Section: Introductionmentioning

confidence: 99%

“…It usually explores the relationship between the per-frame mel-frequency cepstral coefficients (MFCCs) and pitch/energy contours by directly concatenating them into a single vector stream to build a single model. Score-domain fusion [1][2] is the most popular and successful strategy. But it often ignores the dependency between prosodic and spectral cues and independently establishes one system for one information source.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Joint prosodic and spectral modeling for robust speaker verification

Liao,

Chang,

Xie

et al. 2008

Speech Prosody 2008

View full text Add to dashboard Cite

In this paper, a joint prosodic and spectral modeling framework is proposed instead of traditional score-domain fusion approaches to alleviate the problem of mismatch channel/handset/ambient noise. The basic idea is to embed the concept of hierarchical structure of speech prosody into an ergodic HMM (EHMM), and model the prosodic status transitions and prosodic/spectral features by EHMM's states, state transition probabilities and state-dependent observation distributions, respectively. Experimental results evaluated on the standard single-speaker detection task of NIST 2001 speaker recognition evaluation (NIST-SRE 2001) showed that the proposed approach not only outperformed the spectral feature-based baseline (8.04% vs. 8.64% in equal error rate, EER) but also worked a little bit better than score-domain fusion ( 8.44%) approach.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Joint prosodic and spectral modeling for robust speaker verification

Liao,

Chang,

Xie

et al. 2008

Speech Prosody 2008

View full text Add to dashboard Cite

show abstract

“…The most important issue for speaker verification is the channel/handset mismatch problem. To address this problem, higher level information including prosodic cues [1,2] and mode of glottal phonation (or voice-quality) [3] of a speaker, which may be less sensitive to channel/handset mismatch are attractive recently.…”

Section: Introductionmentioning

confidence: 99%

“…The prosodic information, such as the dynamic of pitch/energy contour, lengthening and pause duration, are already known to be informative and complemented with the spectral features-based speaker recognition approaches [1,2]. On the other hand, there are only few papers working on applying the voicequality, especially the normalized amplitude quotient (NAQ) [4], to speaker recognition task so far.…”

Section: Introductionmentioning

confidence: 99%

“…Moreover, the latent prosody analysis (LPA) approach previous proposed in [2] is modified to explore the long-span correlation of successive prosody states (status). The basic idea is to automatically label enrollment utterances of speakers into sequences of prosody keywords (states) and to use chunks of prosody keywords to establish bi-gram speaker models or prosody keyword-speaker co-occurrence matrix.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Exploiting glottal and prosodic information for robust speaker verification

Liao,

Zeng,

Chen

et al. 2006

Speech Prosody 2006

View full text Add to dashboard Cite

In this paper, three different levels of speaker cues including the glottal, prosodic and spectral information are integrated together to build a robust speaker verification system. The major purpose is to resist the distortion of channels and handsets. Especially, the dynamic behavior of normalized amplitude quotient (NAQ) and prosodic feature contours are modeled using Gaussian of mixture models (GMMs) and two latent prosody analyses (LPAs)-based approaches, respectively. The proposed methods are evaluated on the standard one speaker detection task of the 2001 NIST Speaker Recognition Evaluation Corpus where only one 2-minute training and 30-second trial speech (in average) are available. Experimental results have shown that the proposed approach could improve the equal error rates (EERs) of maximum a priori-adapted (MAP)-GMMs and GMMs+T-norm approaches from 12.4% and 9.5% to 10.3% and 8.3% and finally to 7.8%, respectively.

show abstract