양자 간 대화 상황에서의 화자인식을 위한 문장 시퀀싱 방법을 통한 자동 말투 인식

강가람(Garam Kang),권오병(Ohbyung Kwon) 한국지능정보시스템학회 2021 지능정보연구 Vol.27 No.2

Speaker recognition is generally divided into speaker identification and speaker verification. Speaker recognition plays an important function in the automatic voice system, and the importance of speaker recognition technology is becoming more prominent as the recent development of portable devices, voice technology, and audio content fields continue to expand. Previous speaker recognition studies have been conducted with the goal of automatically determining who the speaker is based on voice files and improving accuracy. Speech is an important sociolinguistic subject, and it contains very useful information that reveals the speakers attitude, conversation intention, and personality, and this can be an important clue to speaker recognition. The final ending used in the speakers speech determines the type of sentence or has functions and information such as the speakers intention, psychological attitude, or relationship to the listener. The use of the terminating ending has various probabilities depending on the characteristics of the speaker, so the type and distribution of the terminating ending of a specific unidentified speaker will be helpful in recognizing the speaker. However, there have been few studies that considered speech in the existing text-based speaker recognition, and if speech information is added to the speech signal-based speaker recognition technique, the accuracy of speaker recognition can be further improved. Hence, the purpose of this paper is to propose a novel method using speech style expressed as a sentence-final ending to improve the accuracy of Korean speaker recognition. To this end, a method called sentence sequencing that generates vector values by using the type and frequency of the sentence-final ending appearing in the utterance of a specific person is proposed. To evaluate the performance of the proposed method, learning and performance evaluation were conducted with a actual drama script. The method proposed in this study can be used as a means to improve the performance of Korean speech recognition service.

Implementation of Real-time Wheel Order Recognition System Based on the Predictive Parameters for Speaker’s Intention

Serng-Bae Moon,Seung-Hwan Jun 한국항해항만학회 2011 한국항해항만학회지 Vol.35 No.7

In this paper new enhanced post-process predicting the speaker’s intention was suggested to implement the real-time control module for ship’s autopilot using speech recognition algorithm. The parameter was developed to predict the likeliest wheel order based on the previous order and expected to increase the recognition rate more than pre-recognition process depending on the universal speech recognition algorithms. The values of parameter were assessed by five certified deck officers being good at conning vessel. And the entire wheel order recognition process were programmed to TMS320C5416 DSP so that the system could recognize the speaker’s orders and control the autopilot in real-time. We conducted some experiments to verify the usefulness of suggested module. As a result, we have confirmed that the post-recognition process module could make good enough accuracy in recognition capabilities to realize the autopilot being operated by the speech recognition system.

HMM과 인공신경망을 이용한 화자 적응 연속 음성인식에 관한 연구

김선일 거제전문대학 1998 論文集 Vol.7 No.-

Continuous speech recognition for the RM database is performed for one speaker using HMM. Artificial neural network is implemented to adapt another speakers to this speaker. To do this, DTW is used because there is difference between the lengths of speech data each. New speaker's speech data are used as input for the neural network to have the nonlinear mapping coefficients leaned while the data from which HMMs are teaming are used as targets. The same sentences are used to have the neural network trained. Several combinations of sentences are used. I have chosen several momentums and learning coefficients to find the best values to give maximum recognition rate. Speaker dependent data of RM database are used. For the recognition rate, 42.64% is obtained for the dtd0 speaker when the speech recognition is performed using bef0 HMM models which are obtained by bef0 speaker's data, 70.91% is obtained when the recognition for the dtd0 data being transformed using neural network is performed by bef0 HMM models.

Robust Model Construction Using a Selective Feature Vector for Pattern Recognition with Voice

Jeong-Sik Park,Gil-Jin Jang,Ji-Hwan Kim 보안공학연구지원센터(IJSEIA) 2016 International Journal of Software Engineering and Vol.10 No.1

This paper proposes a new feature vector selection method for voice pattern recognition tasks, especially for speaker or emotion recognition. During the model training phase, robust speaker or emotion models are constructed by using meaningful feature vectors while discarding confusing vectors that may induce recognition error. To select meaningful feature vectors, the proposed method classifies feature vectors into overlapped and non-overlapped sets using log-likelihood ratio. Speaker- and emotion-recognition experiments confirmed that these robust models significantly reduce recognition errors.

〈서경별곡〉의 종합적 고찰

양태순(Yang Tae-Soon) 국어국문학회 2005 국어국문학 Vol.- No.139

The aim of this paper is to examine some controversies over 〈Seogyongbyulgok〉 in terms of its speaker, the division of passages, and interpretation.<br/> 1. The Speaker<br/> There are two theories about the speaker of this poem: one female speaker and multiple speakers theory. The latter contains theories: a different female speakers in each stanza, which emphasizes the work as an organic unity; alternating female and male speakers, which regards each stanza as an independent meaningful unit; and the speaker can both be a male and a female. My opinion supports the one female speaker theory because it is more reasonable to escape the possibility of destroying the sense of unity by introducing multiple speakers.<br/> 2. Division into several passages<br/> There are two theories: One is 3 stanza composition theory, and the other 4 stanza composition theory. I agree to the latter because it is backed up by much more objective and detailed grounds. <br/> 3. The problem of speaker and division is essential to the interpretation of the work under discussion. To me, the first stanza is 'a confession of relative love', the second one 'a confession of absolute love', the third one 'biased grievance', and the fourth 'self recognition'. Especially, the last line of the first stanza "Goesirandae wooreogom jotninoeeda" and the last line of the fourth stanza "Baetadeulmyon Kotkoreeda" need a special attention. The the subject of the subjunctive clause in front of each stanza is 'nim (lover)', and that of the main clause is the poetic speaker. So, the first stanza signifies the speaker's relative love that the speaker will follow the lover, leaving everything behind. The fourth stanza shows a recognition of self immersed in absolute loneliness and lack of meaning of life by showing a resolution to pluck a 'flower' if the lover leaves in a different ship.

Implementation of Real-time Wheel Order Recognition System Based on the Predictive Parameters for Speaker's Intention

문성배,전승환 한국항해항만학회 2011 한국항해항만학회지 Vol.35 No.7

In this paper new enhanced post-process predicting the speaker's intention was suggested to implement the real-time control module for ship's autopilot using speech recognition algorithm. The parameter was developed to predict the likeliest wheel order based on the previous order and expected to increase the recognition rate more than pre-recognition process depending on the universal speech recognition algorithms. The values of parameter were assessed by five certified deck officers being good at conning vessel. And the entire wheel order recognition process were programmed to TMS320C5416 DSP so that the system could recognize the speaker's orders and control the autopilot in real-time. We conducted some experiments to verify the usefulness of suggested module. As a result, we have confirmed that the post-recognition process module could make good enough accuracy in recognition capabilities to realize the autopilot being operated by the speech recognition system.

감정 적응을 이용한 감정 화자 인식

김원구(Weon-Goo Kim) 대한전기학회 2017 전기학회논문지 Vol.66 No.7

Speech with various emotions degrades the performance of the speaker recognition system. In this paper, a speaker recognition method using emotional adaptation has been proposed to improve the performance of speaker recognition system using affective speech. For emotional adaptation, emotional speaker model was generated from speaker model without emotion using a small number of training affective speech and speaker adaptation method. Since it is not easy to obtain a sufficient affective speech for training from a speaker, it is very practical to use a small number of affective speeches in a real situation. The proposed method was evaluated using a Korean database containing four emotions. Experimental results show that the proposed method has better performance than conventional methods in speaker verification and speaker recognition.

DMS 기법을 이용한 화자 독립 단독어 음성 인식

이강성 광운대학교 신기술연구소 1997 신기술연구소논문집 Vol.26 No.-

화자 독립 음성 인식을 위한 DMS방법을 제안한다. DMS (Dynamic Multisection Vector Quantization) 기법이 이미 제안되었으나 화자 종속에 대한 평가만이 이루어져 있다. 새 알고리즘은 여러 화자가 발성한 단어를 집단화 기법을 통해서 몇 개의 집단으로 분류하고, 그 집단 내에서 모델을 작성한다. 실험을 통해서 화자 독립 실험에서 98.0%의 인식율을 얻었다. 이 기법은 학습 횟수가 적은 화자 종속 인식에서부터 학습 횟수가 많은 화자 독립까지 함께 사용하는 방법이다. A new DMS(Dynamic Multisection Vector Quantization) method was proposed for speaker independent isolated-word speech recognition. Although DMS method has been proposed, evaluation for the algorithm was performed only for speaker dependent speech recognition. The proposed algorithm classifies all the words recorded into several clusters and then makes models from each cluster. 98% recognition rate was yielded through the speaker independent recognition experiments. This method can be used in both speaker dependent speech recognition systems which require few recordings for training, and in speaker independent speech recognition systems which require many recordings.

Implementation of Real-time Wheel Order Recognition System Based on the Predictive Parameters for Speaker's Intention

Moon, Serng-Bae,Jun, Seung-Hwan Korean Institute of Navigation and Port Research 2011 한국항해항만학회지 Vol.35 No.7

In this paper new enhanced post-process predicting the speaker's intention was suggested to implement the real-time control module for ship's autopilot using speech recognition algorithm. The parameter was developed to predict the likeliest wheel order based on the previous order and expected to increase the recognition rate more than pre-recognition process depending on the universal speech recognition algorithms. The values of parameter were assessed by five certified deck officers being good at conning vessel. And the entire wheel order recognition process were programmed to TMS320C5416 DSP so that the system could recognize the speaker's orders and control the autopilot in real-time. We conducted some experiments to verify the usefulness of suggested module. As a result, we have confirmed that the post-recognition process module could make good enough accuracy in recognition capabilities to realize the autopilot being operated by the speech recognition system.

MCE기반의 다중 특징 파라미터 스코어의 결합을 통한 화자인식 성능 향상

강지훈,김보람,김규영,이상훈 한국산학기술학회 2020 한국산학기술학회논문지 Vol.21 No.6

본 논문에서는 화자인식 성능 향상을 위해 음원에서 개선된 특징추출 방식과 최소 분류 오차 기반의 다중 특징 벡터 스코어에 대한 가중치 추정을 사용하여 스코어 결합을 제안하였다. 제안한 특징 벡터는 Glottal Flow에서 무의미한 정보구간인 평탄한 스펙트럼 구간을 제거하기 위하여 저역통과 필터를 수행한 신호에서 인지적 선형 예측 캡스트럼 계수, 왜도, 첨도를 추출하여 구성하였다. 제안한 특징 벡터는 종래의 음원에서 멜-주파수 캡스트럼 계수, 인지적 선형 예측 캡스트럼 계수를 추출하여 가우시안 혼합 모델로 모델링한 화자인식 시스템을 개선하기 위해 사용된다. 또한, 스코어 추정과정의 신뢰성을 높이기 위하여 기존의 스코어의 확률 분포를 사용하여 가중치를 추정하는 대신 제안한 특징 벡터에서 평가된 점수와 종래의 특징 벡터에서 평가된 점수에 대하여 최소 분류 오차 기법으로 가중치를 추정하여 스코어를 결합함으로써 최적의 화자를 찾는다. 실험 결과 제안한 특징 벡터가 화자를 인식하는데 유효한 정보를 포함하고 있는 것을 확인하였다. 또한, 최소 분류 오차 기반의 다중 특징 파라미터 스코어를 결합하여 화자인식을 수행하였을 때, 종래의 화자인식 성능보다 더 우수한 성능을 나타내는 것을 확인할 수 있으며, 특히 가우시안 혼합 모델이 낮을 때 더 높은 성능향상을 보였다. In this thesis, an enhanced method for the feature extraction of vocal source signals and score combination using an MCE-Based weight estimation of the score of multiple feature vectors are proposed for the performance improvement of speaker recognition systems. The proposed feature vector is composed of perceptual linear predictive cepstral coefficients, skewness, and kurtosis extracted with lowpass filtered glottal flow signals to eliminate the flat spectrum region, which is a meaningless information section. The proposed feature was used to improve the conventional speaker recognition system utilizing the mel-frequency cepstral coefficients and the perceptual linear predictive cepstral coefficients extracted with the speech signals and Gaussian mixture models. In addition, to increase the reliability of the estimated scores, instead of estimating the weight using the probability distribution of the convectional score, the scores evaluated by the conventional vocal tract, and the proposed feature are fused by the MCE-Based score combination method to find the optimal speaker. The experimental results showed that the proposed feature vectors contained valid information to recognize the speaker. In addition, when speaker recognition is performed by combining the MCE-based multiple feature parameter scores, the recognition system outperformed the conventional one, particularly in low Gaussian mixture cases.