Workshop Homepage
Workshop 2000

The mcs Paper Club

Everyday (well, almost) the mcs group meets to discuss a technical paper relevant to Mandarin Pronunciation Modeling.
Here are the proceedings...
Date: 31/7/2000
Discussion Leader: Liu Yi.
Paper under discussion : Liu Yi & Pascale Fung ,"Modeling Pronunciation Variations in Spontaneous Mandarin Speech"
Date: 1/8/2000
Discussion Leader: Veera Venkataramani
Paper under discussion : S GAO, T LEE, YW Wong, Bo XU, PC Ching & Taiyi HUANG, "Acoustic Modeling for Chinese Speech Recognition: A Comparative Study of Mandarin and Cantonese"
Summary: Comparison of decision tree clustering for context dependent Initial/Final systems for Cantonese and Mandarin read speech.
Key Points: Mandarin is easier to recognize than Cantonese given equal quantities of the 863 and CUSENT data. Provides an analysis of questions used in triphone clustering, and also an analysis of triphone coverage.

Date: 3/8/2000
Discussion Leader: Zhanjiang Song
Paper under discussion: Sung-Chien Lin, Chi-Lung Tsai, et al. Chinese Language Model Adaptation based on Document Classification and Multiple Domain-Specific Language Models, Eurospeech '97.
Key Points:
  • Preparation: Documents Classfication Instead of using convetional Keyword-Class Matrix, it is based on homogeneity of documents: Perplexity (PP) and Word Bigram Coverage (Cwb) with respect to domains.
  • Training: Multiple domain-specific LM, interpolated with a general-domain LM.
  • Recognizing: Dynamically select domain-specific LM.
    Date: 4/8/2000
    Discussion Leader Umar Ruhi
    Paper under discussion : C. J. Chen, R. A. Gopinath, M.D. Monkowski, M. A. Picheny and K. Shen, EuroSpeech97, New Methods in Continuous Mandarin Speech Recognition
    Key Points: Use filtered pitch as a feature, similar to energy. Tone is used at 'phone level' - don't use IFs. Observe that low tone is more 'flexible' in speech than other ones; Liu Yi suggests that this agrees with intuition/experience.
    Date: 7/8/2000
    Discussion Leader Terri Kamm
    Papers under discussion :
  • Sharlene Liu, Sean Doyle, Allen Morris, Farzad Ehasani, ICSLP '98, The Effect of Fundamental Frequency on Mandarin Speech Recognition.
  • Hank Chang-Han Huang and Frank Seide, ICASSP 2000, Pitch Tracking and Tone Features for Mandarin Speech Recognition.
  • Yang Cao, Yonggang Deng, Hong Zhang, Taiyi Huang, Bo Xu, ICASSP 2000, Decision Tree Based Mandarin Tone Model and its Application of Speech Recognition.
    Date: 8/8/2000
    Discussion Leader: Bill Byrne
    Papers under discussion: Michael Riley et. al, Stochastic pronunciation modelling from hand-labelled phonetic corpora.
    Date: 10/8/2000
    Discussion Leader: Zheng, Tom
    Papers under discussion: Trym Hotler and Torbjorn Svendsen, Maximum Likelihood modelling of pronunciation variation. ,Speech Communication, Vol 29, 1999, pp 177-191.

    Veera Venkataramani
    Last modified: Mon Aug 21 13:27:48 EDT 2000