Content-aware local variability vector for speaker verification with short utterance

Chen, Liping; Lee, Kong Aik; Chng, Eng Siong; Ma, Bin; Li, Haizhou; Dai, Li

doi:10.1109/icassp.2016.7472726

Cited by 6 publications

(9 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Speaker verification performs almost ideally in neutral talking environment, while it performs poorly in emotional talking environments. There are many studies that study "speaker verification in neutral environment" [2][3][4][5][6], while few studies spotlight on "speaker verification in emotional environments" [7][8][9][10][11].…”

Section: Prior Workmentioning

confidence: 99%

“…There are many research [2][3][4][5][6] that study speaker verification in neutral environments. The authors of [2] aimed in one of their work at addressing the long-term speaker variability problem in the feature domain in which they extracted more exact speaker-specific and time-insensitive information.…”

Section: Prior Workmentioning

confidence: 99%

“…Their strategy was based on the F-ratio criterion to decide the whole discrimination-sensitivity of frequency bands 4 by including both the session-specific variability data and the speaker-specific information [2]. The authors of [3] proposed extracting local session variability vectors on distinct phonetic classes from the utterances instead of estimating the session variability across the overall utterance as i-vector does. Based on the deep neural network (DNN), the posteriors trained for phone state categorization, local vectors express the session variability contained in specific phonetic content.…”

Section: Prior Workmentioning

confidence: 99%

“…In this work, we address the issue of improving "speaker verification performance in emotional environments" based on proposing, applying, and evaluating a three-stage speaker verification 3 architecture that is consists of three cascaded stages: "gender identification stage" followed by an "emotion identification stage" followed by a "speaker verification stage".…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Three-stage speaker verification architecture in emotional talking environments

Shahin

Nassif

2018

Int J Speech Technol

View full text Add to dashboard Cite

Speaker verification performance in neutral talking environment is usually high, while it is sharply decreased in emotional talking environments. This performance degradation in emotional environments is due to the problem of mismatch between training in neutral environment while testing in emotional environments. In this work, a three-stage speaker verification architecture has been proposed to enhance speaker verification performance in emotional environments. This architecture is comprised of three cascaded stages: gender identification stage followed by an emotion identification stage followed by a speaker verification stage. The proposed framework has been evaluated on two distinct and independent emotional speech datasets: in-house dataset and "Emotional Prosody Speech and Transcripts" dataset. Our results show that speaker verification based on both gender information and emotion information is superior to each of speaker verification based on gender information only, emotion information only, and neither gender information nor emotion information. The attained average speaker verification performance based on the proposed framework is very alike to that attained in subjective assessment by human listeners. 2

show abstract

Section: Prior Workmentioning

confidence: 99%

Section: Prior Workmentioning

confidence: 99%

Section: Prior Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Three-stage speaker verification architecture in emotional talking environments

Shahin

Nassif

2018

Int J Speech Technol

View full text Add to dashboard Cite

show abstract

“…Test utterances are then scored with subregion models. In Chen et al (2016), the authors use the local session variability vectors estimated from certain phonetic components instead of computing the i-vector from the whole utterance. Phonetic classes are obtained by clustering similar senones (group of triphones with similar acoustic properties) that are estimated from posterior probabilities of a DNN trained for phone state classification.…”

Section: Introductionmentioning

confidence: 99%

Deep neural network based i-vector mapping for speaker verification using short utterances

Guo

Qian

et al. 2018

Speech Communication

View full text Add to dashboard Cite

Text-independent speaker recognition using short utterances is a highly challenging task due to the large variation and content mismatch between short utterances. I-vector and probabilistic linear discriminant analysis (PLDA) based systems have become the standard in speaker verification applications, but they are less effective with short utterances. In this paper, we first compare two state-of-the-art universal background model (UBM) training methods for i-vector modeling using full-length and short utterance evaluation tasks. The two methods are Gaussian mixture model (GMM) based (denoted I-vector GMM) and deep neural network (DNN) based (denoted as I-vector DNN) methods. The results indicate that the I-vector DNN system outperforms the I-vector GMM system under various durations (from full length to 5 s). However, the performances of both systems degrade significantly as the duration of the utterances decreases. To address this issue, we propose two novel nonlinear mapping methods which train DNN models to map the i-vectors extracted from short utterances to their corresponding long-utterance i-vectors. The mapped i-vector can restore missing information and reduce the variance of the original short-utterance i-vectors. The proposed methods both model the joint representation of short and long utterance i-vectors: the first method trains an autoencoder first using concatenated short and long utterance i-vectors and then uses the pre-trained weights to initialize a supervised regression model from the short to long version; the second method jointly trains the supervised regression model with an autoencoder reconstructing the short utterance i-vector itself. Experimental results using the NIST SRE 2010 dataset show that both methods provide significant improvement and result in a 24.51% relative improvement in Equal Error Rates (EERs) from a baseline system. In order to learn a better joint representation, we further investigate the effect of a deep encoder with residual blocks, and the results indicate that the residual network can further improve the EERs of a baseline system by up to 26.47%. Moreover, in order to improve the short i-vector mapping to its long version, an additional vector, which represents the average value of phoneme posteriors across frames, is also added to the input, and results in a 28.43% improvement. When further testing the best-validated models of SRE10 on the Speaker In The Wild (SITW) dataset, the methods result in a 23.12% improvement on arbitrary-duration (1-5 s) short-utterance conditions.

show abstract

Speaker-Phonetic Vector Estimation for Short Duration Speaker Verification

Sethu

Ambikairajah

et al. 2018

2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Content-aware local variability vector for speaker verification with short utterance

Cited by 6 publications

References 12 publications

Three-stage speaker verification architecture in emotional talking environments

Three-stage speaker verification architecture in emotional talking environments

Deep neural network based i-vector mapping for speaker verification using short utterances

Speaker-Phonetic Vector Estimation for Short Duration Speaker Verification

Contact Info

Product

Resources

About