Comparison of text-independent speaker recognition methods using VQ-distortion and discrete/continuous HMMs

Matsui, Tomoko; Furui, Sadaoki

doi:10.1109/icassp.1992.226096

Cited by 119 publications

(61 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…(6) that determines which features are outliers. In our implementation a value α = 3 was employed that also coincides with the value used by Matsui and Furui (1992). The limit imposed by this value for α does not degrade the performance of the system in absence of outliers, since, as it is well known, for a Gaussian distribution the 99.7 % of the samples fall within ±3σ (the experimental results that show the good performance of the method in clean conditions are given in Table 2, Section 4).…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Matsui and Furui (1991) use a new distance called DIM (Distortion-Intersection Measure) in a vector quantisation-based speaker recognition system. Later, this distance was adapted to a HMM-based speaker recogniser (Matsui and Furui (1992)). …”

Section: Motivation and Antecedentsmentioning

confidence: 99%

“…In this paper we suggest a method that we call bounded-distance HMM, which resembles a method already proposed in speaker recognition (Matsui and Furui (1992)). The aim of bounded-distance HMM is to mitigate the influence of the features that are outliers for each acoustic model.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

The synergy between bounded-distance HMM and spectral subtraction for robust speech recognition

Vicente-Peña

Díaz-de-María

Kleijn

2010

Speech Communication

View full text Add to dashboard Cite

Additive noise generates important losses in automatic speech recognition systems. In this paper, we show that one of the causes contributing to these losses is the fact that conventional recognisers take into consideration feature values that are outliers. The method that we call bounded-distance HMM is a suitable method to avoid that outliers contribute to the recogniser decision. However, this method just deals with outliers, leaving the remaining features unaltered. In contrast, spectral subtraction is able to correct all the features at the expense of introducing some artifacts that, as shown in the paper, cause a larger number of outliers. As a result, we find that bounded-distance HMM and spectral subtraction complement each other well. A comprehensive experimental evaluation was conducted, considering several well-known ASR tasks (of different complexities) and numerous noise types and SNRs. The achieved results show that the suggested combination generally outperforms both the bounded-distance HMM and spectral subtraction individually. Furthermore, the obtained improvements, especially for low and medium SNRs, are larger than the sum of the improvements individually obtained by bounded-distance HMM and spectral subtraction.

show abstract

Section: Implementation Detailsmentioning

confidence: 99%

Section: Motivation and Antecedentsmentioning

confidence: 99%

See 1 more Smart Citation

The synergy between bounded-distance HMM and spectral subtraction for robust speech recognition

Vicente-Peña

Díaz-de-María

Kleijn

2010

Speech Communication

View full text Add to dashboard Cite

show abstract

“…Figure 5: Recognition rate and a combination factor for the dierent combinations. 5. CONCLUSION In this paper we h a v e proposed a speaker recognition model using Two-Dimensional Mel-Cepstrum and a predictive network.…”

Section: Robustness For Time Variationmentioning

confidence: 99%

“…Furthermore, the VQ-based methods using a codebook for each speaker and the HMM-based methods have been reported to been robust against utterance variation and give good a performance. However, it has been reported that when the training data is small, the performance of the VQ method decreases less than the HMM method [5]. This paper describes a speaker recognition model using a static feature model and dynamic feature model.…”

Section: Introductionmentioning

confidence: 99%

Speaker recognition model using two-dimensional mel-cepstrum and predictive neural network

Kitamura

Takei²

Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96

View full text Add to dashboard Cite

This paper describes a speaker recognition model using TwoDimensional Mel-Cepstrum and predictive neural network. The speaker model consists of two networks. The rst one is a self-organizing VQ map(Kohonen's feature map). The second part is the predictive network and learns transitional patterns on the feature map of each speaker's model. TDMC consists of averaged features and dynamic features of the two-dimensional mel-log spectra in the analyzed interval. The measure for speaker recognition is obtained by using a combination of the VQ distortion on the feature map and the prediction error on the predictive network. In the study, text-independent speaker identication experiments for 8 speakers were carried out. The experimental results have shown that a combination of a feature map and a predictive network is very eective, and that the proposed model using TDMC shows the robustness for time interval.

show abstract

Spoken language identification by ergodic HMMS and its state sequences

Nakagawa

Seino

Ueda

1994

Electron Comm Jpn Pt III

View full text Add to dashboard Cite

This paper describes an automatic text‐ and speaker‐independent language identification method based on hidden Markov models (HMMs) for acoustic features. The hidden Markov modeling is used to represent the phonotactics for each language. Each language has its proper phonotactics. The HMM topology here is a fully structured (ergodic) model wherein any state could transit to all states. Two kinds of HMMs are used: the discrete HMM (DHMM) with the codebook and the continuous density HMM (CHMM). The HMM was trained using both the Baum‐Welch (forward‐backward) algorithm and the Viterbi algorithm. The latter was used for emphasizing the state transition probability. For comparison, experiments also were conducted on the identification using a mixtured Gaussian distribution model with one state. This single‐state Gaussian distribution model gave the same performance as the HMM trained with the Baum‐Welch algorithm. This is because the transition between states which reflects the characteristics of each language does not affect the likelihood scores very much. This problem was addressed by emphasizing the transition probabilities and using the Viterbi algorithm, which resulted in an improvement in the recognition rates. The trigram for optimal state sequence is introduced. Combining it with the HMM produced the best results.

show abstract

Comparison of text-independent speaker recognition methods using VQ-distortion and discrete/continuous HMMs

Cited by 119 publications

References 8 publications

The synergy between bounded-distance HMM and spectral subtraction for robust speech recognition

The synergy between bounded-distance HMM and spectral subtraction for robust speech recognition

Speaker recognition model using two-dimensional mel-cepstrum and predictive neural network

Spoken language identification by ergodic HMMS and its state sequences

Contact Info

Product

Resources

About