Unsupervised and active learning in automatic speech recognition for call classification

Published 28 September 2004

Dilek Hakkani‐Tür, Gökhan Tür, M. Rahim, Giuseppe Riccardi

DOI10.1109/icassp.2004.1326014

Citations41

TL;DR

A novel approach that aims at reducing the amount of manually transcribed in-domain data required for building automatic speech recognition (ASR) models in spoken language dialog systems based on mining relevant text from various conversational systems and Web sites is presented.

Abstract

A key challenge in rapidly building spoken natural language dialog applications is minimizing the manual effort required in transcribing and labeling speech data. This task is not only expensive but also time consuming. We present a novel approach that aims at reducing the amount of manually transcribed in-domain data required for building automatic speech recognition (ASR) models in spoken language dialog systems. Our method is based on mining relevant text from various conversational systems and Web sites. An iterative process is employed where the performance of the models can be improved through both unsupervised and active learning of the ASR models. We have evaluated the robustness of our approach on a call classification task that has been selected from AT&T VoiceTone/sup SM/ customer care. Our results indicate that with unsupervised learning it is possible to achieve a call classification performance that is only 1.5% lower than the upper bound set when using all available in-domain transcribed data.

Keywords

Computer Science

Machine LearningBoosTexter: A Boosting-based System for Text Categorization

2,194 Citations2000Robert E. Schapire, Yoram Singer

This work describes in detail an implementation, called BoosTexter, of the new boosting algorithms for text categorization tasks, and presents results comparing the performance of Boos Texter and a number of other text-categorization algorithms on a variety of tasks.

View PDF

IEEE Transactions on Speech and Audio ProcessingMaximum a posteriori estimation for multivariate Gaussian mixture observations of Markov chains

2,102 Citations1994J.-L. Gauvain, Chin‐Hui Lee

A framework for maximum a posteriori (MAP) estimation of hidden Markov models (HMM) is presented, and Bayesian learning is shown to serve as a unified approach for a wide range of speech recognition applications.

Speech CommunicationHow may I help you?

507 Citations1997Allen L. Gorin, Giuseppe Riccardi +1 more

This paper focuses on the task of automatically routing telephone calls based on a user's fluently spoken response to the open-ended prompt of “ How may I help you? ”.

2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03).Unsupervised language model adaptation

136 Citations2003Michiel Bacchiani, Brian Roark

Unsupervised language model adaptation, from ASR transcripts, shows an error rate reduction of 3.9% over the unadapted baseline performance, from 28% to 24.1%, using 17 hours of unsupervised adaptation material.

2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03).A general algorithm for word graph matrix decomposition

46 Citations2003Dilek Hakkani‐Tür, Giuseppe Riccardi

A general framework for lattice chunking, the pivot algorithm, which has applications to many tasks (e.g. parsing, named entity extraction) and results on the performance of confidence scores for different large vocabulary speech recognition tasks are presented.

Effective utterance classification with unsupervised phonotactic models

31 Citations2003Hiyan Alshawi

The method combines domain independent acoustic models with off-the-shelf classifiers to give utterance classification performance that is surprisingly close to what can be achieved using conventional word-trigram recognition requiring manual transcription.

View PDF

Language model adaptation via minimum discrimination information

19 Citations2002P. Srinivasa Rao, Michael D. Monkowski +1 more

The authors first discuss the adaptation of general trigram language models to a known topic using the minimum discrimination information (MDI) method, and present results on the switchboard corpus which consists of telephone conversations on several topics.

IEEE International Conference on Acoustics Speech and Signal ProcessingUnsupervised training techniques for natural language call routing

9 Citations2002Rukmini Iyer, H. Gish +1 more

The authors' state-of-the-art speech recognition technology is leveraged and automatic transcriptions are used to reduce time and manual effort in developing a call routing application.