Approaches to topic identification on the switchboard corpus

McDonough, John; Ng, Kenney; Jeanrenaud, P.; Gish, H.; Rohlíček, Ján

doi:10.1109/icassp.1994.389275

Cited by 49 publications

(26 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Keyword likelihoods, weighted by learned sigmoid functions, were used as input to the message classifier. In [181], an "expected number of occurrences" of keywords in messages is obtained by summing the scores of keyword hypotheses that are generated by a word spotter. Thresholding is a common practice for keyword spotting systems and also for word features derived from lattice scans as in [122,132].…”

Section: Integrating Confidence Measures Into Ir Modelsmentioning

confidence: 99%

See 1 more Smart Citation

Automatic Summarization

Larson

2012

FNT in Information Retrieval

View full text Add to dashboard Cite

Section: Integrating Confidence Measures Into Ir Modelsmentioning

confidence: 99%

“…The early speech retrieval systems, in particular [131,181,231], used word-level features derived by a word spotter in order to perform classification and retrieval of spoken documents. A word spotter takes as input a list of words and the speech signal and outputs points in time at which words appear.…”

Section: Hybrid Approaches For Scrmentioning

confidence: 99%

Automatic Summarization

Larson

2012

FNT in Information Retrieval

View full text Add to dashboard Cite

“…Topic identification based on recorded speech was initially investigated in connection with call routing [12] and classification of audio messages on the switchboard corpus [25]. More recently, several variants of topic identification methods have been applied to the analysis of meeting data.…”

Section: Related Workmentioning

confidence: 99%

Classification of patient case discussions through analysis of vocalisation graphs

Luz

Kane

2009

Proceedings of the 2009 International Conference on Multimodal Interfaces

View full text Add to dashboard Cite

This paper investigates the use of amount and structure of talk as a basis for automatic classification of patient case discussions in multidisciplinary medical team meetings recorded in a real-world setting. We model patient case discussions as vocalisation graphs, building on research from the fields of interaction analysis and social psychology. These graphs are "content free" in that they only encode patterns of vocalisation and silence. The fact that it does not rely on automatic transcription makes the technique presented in this paper an attractive complement to more sophisticated speech processing methods as a means of indexing medical team meetings. We show that despite the simplicity of the underlying representation mechanism, accurate classification performance (F-scores: F1 = 0.98, for medical patient case discussions, and F1 = 0.97, for surgical case discussions) can be achieved with a simple k-nearest neighbour classifier when vocalisations are represented at the level of individual speakers. Possible applications of the method in health informatics for storage and retrieval of multimedia medical meeting records are discussed.

show abstract

“…More additional descriptions and techniques can be found in a book chapter by [4]. One of the most common written text classification approach used in TID approach is the naive Bayes approach and it has been borrowed in speech-based systems too ( [5], [6], [7] and [1]). Nevertheless, many of the topic classification methods are applied to written Arabic news and most of the studies and papers rarely discuss TID in spoken Arabic files.…”

Section: Literature Reviewmentioning

confidence: 99%

Spoken Arabic News Classification Based on Speech Features

Abusulaiman¹

2017

IJRASET

View full text Add to dashboard Cite

One of the most important consequences of what is known as the "Internet era" is the widespread of varied electronic data. This deployment urgently requires an automated system to classify these data to facilitate search and access to the topic in question. This system is commonly used in written texts. Because of the huge increase of spoken files nowadays, there is an acute need for building an automatic system to classify spoken files based on topics. This system has been discussed in the previous researches applied to spoken English texts, but it rarely takes into consideration spoken Arabic texts because Arabic language is challenging and its dataset is rare and not suitable for topic classification. To deal with this challenge, a new dataset is established depending on converting the common written text (ALJ-NEWS) which is widely used in researches in classifying written texts. Then, keywords extraction method is implemented in order to extract the keywords representing each class depending on using DTW. Finally, topic identification, based on (MFCC, PLP-RASTA) as speech features and (DTW, HMM) as identifiers, is created using a technique that is different from the traditional way, using ASR to extract the transcriptions. Regarding the evaluation of the system, F1-measure, precision and recall are used as evaluation metrics. The proposed system shows positive results in the topic classification field. The F1-measure for topic identification system using DTW classifier records 90.26% and 91.36% using HMM classifier in the average. In addition, the system achieves 89.65% of keywords identification accuracy.

show abstract

Approaches to topic identification on the switchboard corpus

Cited by 49 publications

References 10 publications

Automatic Summarization

Automatic Summarization

Classification of patient case discussions through analysis of vocalisation graphs

Spoken Arabic News Classification Based on Speech Features

Contact Info

Product

Resources

About