Journal of KIISE:Software and Applications (한국정보과학회논문지:소프트웨어및응용)

Korean Institute of Information Scientists and Engineers (한국정보과학회)
권호 표지

Performance Improvement of a Korean Prosodic Phrase Boundary Prediction Model using Efficient Feature Selection

효율적인 기계학습 자질 선별을 통한 한국어 운율구 경계 예측 모델의 성능 향상

  • 김민호 (부산대학교 컴퓨터공학과) ;
  • 권혁철 (부산대학교 정보컴퓨터공학부)
  • Received : 2010.06.04
  • Accepted : 2010.09.27
  • Published : 2010.11.15
Download PDF
⟨ Previous Next ⟩

Abstract

Prediction of the prosodic phrase boundary is one of the most important natural language processing tasks. We propose, for the natural prediction of the Korean prosodic phrase boundary, a statistical approach incorporating efficient learning features. These new features reflect the factors that affect generation of the prosodic phrase boundary better than existing learning features. Notably, moreover, such learning features, extracted according to the hand-crafted prosodic phrase boundary prediction rule, impart higher accuracy. We developed a statistical model for Korean prosodic phrase boundaries based on the proposed new features. The results were 86.63% accuracy for three levels (major break, minor break, no break) and 81.14% accuracy for six levels (major break with falling tone/rising tone, minor break with falling tone/rising tone/middle tone, no break).

운율구 경계 예측은 대화체 음성합성을 실현하기 위한 주요한 자연언어처리 기술 중 하나이다. 본 논문은 자연스러운 한국어 운율구 경계 예측을 실현하고자 기존의 학습 자질을 대신할 새로운 학습 자질을 제안한다. 이 새로운 자질들은 기존의 학습 자질보다 실제 언어생활에서 운율구 경계 발생에 영향을 미치는 여러 요인을 더 잘 반영한다. 특히, 수작업으로 구축한 운율구 경계 예측 규칙을 이용하여 추출한 학습 자질은 높은 정확도 향상에 이바지한다. 본 논문에서 제안한 새로운 학습 자질을 바탕으로 CRFs(Conditional Random Fields)를 이용하여 운율구 경계 예측 모델을 만들었다. 그 결과 3단계 운율구 경계(강한 경계, 약한 경계, 운율구 내부 비경계) 예측에서 86.63%의 정확도를, 6단계 운율구 경계(상승조/하강조 강한 경계, 상승조/하강조/평탄조 약한 경계, 운율구 내부 비경계) 예측에서는 81.14%의 정확도를 보였다.

Keywords

References

  1. Taylor, P., Black, A. W., "Assigning Phrase Breaks from Part-of-Speech Sequences," In Proceedings of Eurospeech, pp.995-998, 1997.
  2. Lee, S., Oh, Y., "Tree-based modeling of prosodic phrasing and segmental duration for Korean TTS systems," Speech Communication, vol.28, pp.283-300, 1999. https://doi.org/10.1016/S0167-6393(99)00014-X
  3. Kim, B., Lee, G., G., "Implementation of Korean TTS System based on Natural Language Processing," Malsori, vol.46, pp.51-64, 2003. (in Korean)
  4. Jeong, H., Study on Korean Nouns, Hangugmunhwasa, 2002. (in Korean)
  5. Kim, H., "The Construction of Adverb Lexicon in Contemporary Korean - On Some Issues of the description and the Classification of Adverbs -," Korean Journal of Linguistics, vol.24, pp.109-144, 1999. (in Korean)
  6. Kwon, J., Kim, Y., Moon, Y., et al., "A Study on the Interface between Syntactic and Prosodic Structure with Special Reference to the Modes of Ambiguity Resolution," Korean Journal of Linguistics, vol.20, pp.103-109, 1997. (in Korean)
  7. Kim, S., Rhythmic Units and Syntactic Structures in Korean: A Phonetic and Linguistic Study Aiming at Improving the Rhythmic Properties of Synthetic Speech, Seoul National University, 2002. (in Korean)
  8. Lee, Chan-Do, "A Computation Study of Prosodic Structures of Korean for Speech Recognition and Synthesis: Predicting Phonological Boundaries," The Transactions of the Korea Information Processing Society, vol.4, no.1, pp.280-287, 1997. (in Korean)
  9. Hirschberg, J., Prieto, P., "Training International Phrasing Rules Automatically for English and Spanish Text-to-Speech," Speech Communication, vol.18, pp.281-290, 1996. https://doi.org/10.1016/0167-6393(96)00017-9
  10. Ostendorf, M., Veilleus, N., "A Hierarchical Stochastic Model for Automatic Prediction of Prosodic Boundary Location," Computational Linguistics, vol.20, no.1, pp.27-54, 1994.
  11. Kim, S., Kim, B., Jeoung, M., Lee, G., "Using CRF (Contional Random Fields) to Predict," Human & Cognitive Language Technology 2005, pp.134-138, 2005. (in Korean)
  12. Yarowsky, D., "Homograph Disambiguation in Text-to-speech Synthesis," Progress in Speech Synthesis, pp.366-377, 1996.
  13. Jung, Y., Cho, S., Yoon, A., Kwon, H.-C., "Prediction of Prosodic Break Using Syntatic Relations and Prosodic Features," Korean Journal of Cognitive Science, vol.19, no.1, pp.89-105, 2008. (in Korean)
  14. Lee, S., Oh, Y.-H., "The Modeling of Prosodic Phrasing and Pause Duration using CART," Korean Scientific and Cultural Studies Programme Workshop 1998, vol.15, no.1, pp.81-86, 1998. (in Korean)
  15. Chun, Jin.-w., Kim, H. W., Kim, D. g., Lee, Y., "Prosodic-Boundaray Prediction for Korean Textto- Speech System," In Proceedings of Acoustical Society of Korea, vol.22, no.1, pp.77-83, 2002. (in Korean)
  16. Ostendorf, M., Veilleus, N. "A hierarchical Stochastic Model for Automatic prediction of Prosodic Boundary Location," Computational Linguistics, vol.20, no.1, pp.27-54, 1994.
  17. J Lafferty, A McCallum, F Pereira, "Conditional random fields: Probabilistic models for segmenting and labeling sequence data," Machine Learning- International Workshop then Conference, 2001.
  18. Jung, I., Reliable Prediction of Prosodic Breaks by Combining Rules and probabilities Obtained from Small-Scale Corpus, Pusan National University, 2009.