• Publications
  • Influence
A direct LDA algorithm for high-dimensional data - with application to face recognition
However, for a task with very high dimensional data such as images, the traditional LDA algorithm encounters several difficulties, and before LDA can be used to reduce dimensionality, another procedure has to be first applied for dimensionality reduction. Expand
The ISL meeting corpus: the impact of meeting type on speech style
An experiment is described which aimed at testing the possibility of controlling issues of speaking style by meeting type and results show that depending on the meeting type, speaking style varies in terms of turn length, speed and disfluencies. Expand
Advances in automatic meeting record creation and access
The paper summarizes part of the progress that has been made in this test bed on the question of automatic transcription using large vocabulary continuous speech recognition, information access using non-keyword based methods, summarization and user interfaces. Expand
Enhanced tree clustering with single pronunciation dictionary for conversational speech recognition
It is shown that accurate pronunciation modeling can be achieved through efficient parameter sharing in the acoustic model, and a 1.8% absolute word error rate improvement is achieved on Switchboard, a large vocabulary conversational speech recognition task. Expand
Multimodal Meeting Tracker
This paper will examine a meeting room system under development at Carnegie Mellon University that enables us to track, capture and integrate the important aspects of a meeting from people identification to meeting transcription. Expand
New developments in automatic meeting transcription
A prototype system which uses the Web as a source for vocabulary expansion, and preliminary OOV retrieval results are presented, and it is shown how recognition of short turns can be improved by building a language model tailored specifically for short turns. Expand
Automatic speech summarization applied to English broadcast news speech
English broadcast news speech transcribed using a speech recognizer is automatically summarized and a summarization method for multiple utterances using two-level DP technique is proposed. Expand
The ISL Meeting Room System
Oral communication is transient but many important decisions, social contracts and fact findings are first carried out in an oral setup, documented in written form and later retrieved. At CarnegieExpand
Streamlining the front end of a speech recognizer
The invariance property of recognition performance with respect to linear transformations at the front end is studied, which reveals that several LT steps can be consolidated into a single LT, which effectively eliminates the Discrete Cosine Transform step. Expand
The ISL RT04 Mandarin Broadcast News Evaluation System
This paper describes our effort in developing a Mandarin Broadcast News system for the RT-04f (Rich Transcription) evaluation. Starting from a legacy system, we revisited all the issues includingExpand