• 한국통신학회논문지
• /
• 제34권3C호
• /
• pp.257-267
• /
• 2009

This paper presents a training method for neural networks and the employment of MSE (mean scare error) values as the basis of a decision regarding the identity claim of a speaker in a recurrent neural networks based speaker verification system. Recurrent neural networks (RNNs) are employed to capture temporally dynamic characteristics of speech signal. In the process of supervised learning for RNNs, target outputs are automatically generated and the generated target outputs are made to represent the temporal variation of input speech sounds. To increase the capability of discriminating between the true speaker and an impostor, a discriminative training method for RNNs is presented. This paper shows the use and the effectiveness of the MSE value, which is obtained from the Euclidean distance between the target outputs and the outputs of networks for test speech sounds of a speaker, as the basis of speaker verification. In terms of equal error rates, results of experiments, which have been performed using the Korean speech database, show that the proposed speaker verification system exhibits better performance than a conventional hidden Markov model based speaker verification system.

• 대한음성학회지:말소리
• /
• 제47호
• /
• pp.97-107
• /
• 2003

Speaker change detection is to detect automatically a point of time at which speaker was replaced. Since feature parameters used for speaker change detection depend not only on speaker characteristics but also on phonetic characteristics, spoken contents included in the feature parameters inevitably causes performance degradation of speaker change detection. In this paper, to alleviate this problem, a method to normalize phonetic variations in speech feature parameters is proposed for emphasizing changes due to speaker characteristics. Experimental results show that the proposed method improves the performance of speaker change detection.

• 음성과학
• /
• 제8권4호
• /
• pp.59-72
• /
• 2001

In this paper, we experimented with speaker change detection that uses a statistical method for NOD (News On Demand) service. A specified speaker's change can find out content of each data in speech if analysed because it means change of data contents in news data. Speaker change detection acts as preprocessor that divide input speech by speaker. This is an important preprocessor phase for speaker tracking. We detected speaker change using GLR(generalized likelihood ratio) distance base division and BIC (Bayesian information criterion) base division among matrix method. An experiment verified speaker change point using BIC base division after divide by speaker unit using GLR distance base method first. In the experimental result, FAR (False Alarm Rate) was 63.29 in high noise environment and FAR was 54.28 in low noise environment in MDR (Missed Detection Rate) 15% neighborhood.

• 음성과학
• /
• 제7권4호
• /
• pp.41-47
• /
• 2000

This paper concerns an effective text-prompted speaker verification method to increase the performance of speaker verification. While various speaker verification methods have already been developed, their effectiveness has not yet been formally proven in terms of achieving an acceptable performance level. It is also noted that the traditional methods were focused primarily on single, prompted utterance for verification. This paper, instead, proposes sequential decision method using variable threshold focused at handling two utterances for text-prompted speaker verification. Experimental results show that the proposed speaker verification method outperforms that of the speaker verification scheme without using the sequential decision by a factor of up to 3 times. From these results, we show that the proposed method is highly effective and achieves a reliable performance suitable for practical applications.

• 한국음향학회지
• /
• 제31권4호
• /
• pp.266-272
• /
• 2012

화자 분할 기술은 오디오 데이터로부터 자동적으로 화자 경계 구간을 검출하는 것이다. 화자 분할 방식은 화자에 대한 선행 지식 사용 여부에 따라 거리기반 방식과 모델기반 방식으로 나누어진다. 본 논문에서는 eigenvoice 기반의 화자가중치 거리를 이용한 화자 분할 방식을 도입하고, 이 방식을 대표적인 거리 기반 방식들과 비교한다. 또한, 화자가중치의 거리 측정 함수로 유클리드 거리와 cosine 유사도를 사용하여 화자 분할 성능을 비교하고, eigenvoice 방식에 의해 화자 적응된 모델들 사이의 직접적인 거리를 이용한 화자 분할 방식과의 비교를 통해 화자가중치 거리를 이용한 방식이 계산량면에서 효율적인 점을 검증한다.

• 한국음향학회지
• /
• 제40권5호
• /
• pp.466-472
• /
• 2021

화자 전환 검출은 대화 중에 발성 화자가 다른 사람으로 바뀌는 시점을 검출하는 것을 의미한다. 이 과정에서 화자 중복, 화자 정보 표기의 부정확성, 데이터 불균형 등으로 화자가 바뀌는 순간을 검출하는 데 어려움이 발생한다. 본 논문에서는 이러한 문제를 해결하기 위해 음성 인식에 널리 사용되는 TIMIT 데이터를 가공하여 충분한 양의 훈련 데이터를 얻었으며, 화자가 겹치는지를 파악한 후에 화자 전환 여부를 판단하였다. 본 논문에서는 화자 겹침을 고려한 화자 전환 검출 시스템을 구축하기 위하여 다양한 접근법을 사용하여 성능을 평가하고 검증했다. 그 결과 화자 겹칩 영역을 제거하기 위해 X-Vector 구조와 유사한 형태의 검출 시스템과 화자 전환 검출 시스템을 모델링하기 위한 Bi-LSTM 모델을 제안하였다. 실험 결과 기준 시스템보다 상대적으로 각각 4.6 %, 13.8 % 성능 향상을 확인하였다. 또한, 실험 결과를 기반으로 텍스트 정보와 화자 정보 등을 고려한다면 좀 더 강인한 화자 전환 검출 시스템을 구축할 수 있을 것으로 판단한다.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2003년도 10월 학술대회지
• /
• pp.185-188
• /
• 2003

In this paper, we construct phonetic GMM for text-independent speaker identification system. The basic idea is to combine of the advantages of baseline GMM and HMM. GMM is more proper for text-independent speaker identification system. In text-dependent system, HMM do work better. Phonetic GMM represents more sophistgate text-dependent speaker model based on text-independent speaker model. In speaker identification system, phonetic GMM using HMM-based speaker-independent phoneme recognition results in better performance than baseline GMM. In addition to the method, N-best recognition algorithm used to decrease the computation complexity and to be applicable to new speakers.

• 대한음성학회지:말소리
• /
• 제56호
• /
• pp.147-158
• /
• 2005

In this paper, we are focused to develop the simplified speaker verification algorithm without background speaker models, which will be adopted in the portable speaker verification system equipped in portable terminals such as mobile phone and PMP. According to the tolerance interval analysis, the population of someone's speaker model can be represented by a suitable number of selected independent samples of speaker model. So we can make the representative speaker model and threshold under the specified confidence level and coverage. Using proposed algorithm with the number of samples is 40, the experiments show that the false rejection rate is $3.0\%$ and the false acceptance rate $4.3\%$, worth comparing to conventional method's results, $5.4\%\;and\;5.5\%$, respectively. Next step of research will be on the suitable adaptation methods to overcome speech variation problems due to aging effect and operating environments.

• 음성과학
• /
• 제12권4호
• /
• pp.31-42
• /
• 2005

Recently, Gaussian Mixture Model (GMM), a special form of CHMM, has been applied to speaker identification and it has proved that performance of GMM is better than CHMM. Therefore, in this paper the speaker models based on GMM and a new GMM using the modified EM algorithm are introduced and evaluated for text-independent speaker identification. Various experiments were performed to evaluate identification performance of two algorithms. As a result of the experiments, the GMM speaker model attained 94.6% identification accuracy using 40 seconds of training data and 32 mixtures and 97.8% accuracy using 80 seconds of training data and 64 mixtures. On the other hand, the new GMM speaker model achieved 95.0% identification accuracy using 40 seconds of training data and 32 mixtures and 98.2% accuracy using 80 seconds of training data and 64 mixtures. It shows that the new GMM speaker identification performance is better than the GMM speaker identification performance.

• 대한음성학회지:말소리
• /
• 제58호
• /
• pp.83-99
• /
• 2006

One key factor that hinders the widespread deployment of speaker identification technologies is the requirement of long enrollment utterances to guarantee low error rate during identification. To gain user acceptance of speaker identification technologies, adaptation algorithms that can enroll speakers with short utterances are highly essential. To this end, this paper applies MLLR speaker adaptation for speaker enrollment and compares its performance against other speaker modeling techniques: GMMs and HMM. Also, to speed up the computational procedure of identification, we apply speaker clustering method which uses principal component analysis (PCA) and weighted Euclidean distance as distance measurement. Experimental results show that MLLR adapted modeling method is most effective for short enrollment utterances and that the GMMs performs better when long utterances are available.