James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, Speech, Vision and Robotic Group, Cambridge, U.K., February 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the effect of recognition errors and to refine the hst of retrieved messages. Thus, the topic specification and search strategies developed for conventional text based information retrieval (IR) must be adapted to this new environment. This is a challenging problem as indicated in related work [1] [2]. Earher work in the VMR project successfully demonstrated retrieval of spontaneously spoken messages using a small, a priori known set of 35 search keys, for both talker dependent [3] and talker independent [4] word spot ting. This paper describes work using large vocabulary (LV) recognition ....

....model. Furthermore, it has been shown that better retrieval performance can be obtained by combining LV and WS systems, though the best combination method depends somewhat on the nature of the queries. Current work is in progress to augment these two methods with a phone lattice scanning approach [2 ] to cover search keys not in the recognition lexicon or a fixed keyword set, yielding a task independent, truly open vocabulary audio retrieval system. 7. ACKNOWLEDGEMENTS This project is supported by the UK DTI Grant IED4 1 5804 and SERC Grant GR H87629. The authors would like to thank David ....

D. A. James. The Application of Classical Informa- tion Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University, February 1995.

....gives us important information about the nature of the news discourse (for example the optimal segment width) that would be difficult to determine otherwise. Previous work, by the VMR group and by others, has shown that spoken document retrieval using speech recogni tion is becoming practical [8, 7]. Thus the work presented here is only the first step towards general audio and video retrieval; future work in this project will use large vocabulary speech recognition to determine document content. The teletext transcriptions not only serve as a useful benchmark with which to compare future ....

D. A. James. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University, Downing College, February 1995.

....trained on a large corpus of labelled speech data. Given a trained set of HMMs, there exists an efficient algo rithm for finding the most likely model sequence (the recognised words) given unknown speech data. The work presented here takes a different approach, based on the work of James [9]. An off line HMM system is used to generate a number of likely phone sequences, which may then be rapidly searched to find phone strings comprising a desired word. For HMM training and recognition, the acoustic data was parameterised into a spectral represen tation (mel cepstral coefficients) ....

....VMR This is particularly true in the video mail environment, where the vast majority of messages are just talking head images from a small pool of users, against static backgrounds. group and by others, has shown that spoken document re trieval using speech recognition is becoming practical [12, 9]. Future work in this project will be to integrate different audio index sources available from large vocabulary recognition and conventional as well as lattice based word spotting [14, 13] In addition, work will need to be done to make the system robust to environmental noise, microphone ....

D. A. James. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University, February 1995.

....Content extraction here is carried out by following the current state of the art procedures in speech recognition technology. This will involve the use of a fully automated content extraction [35] technique (speech to text conversion) and selected content extraction using word spotting [43, 99], which determines the occurrence of keywords in audio where these keywords are derived from ontologies. In this dissertation we argue that ontology, by reducing the chance of speech recognition error, can provide a means for selected content extraction which will determine which keywords should ....

....data. Therefore, to insure the appropriate selection and presentation of audio information, we advocate selected content extraction. For this process, our goal is to identify a particular set of keywords in the audio segment. For this, the techniques developed in word spotting can be employed [43, 99]. 3.2.1 Word spotting Word spotting techniques can provide selected content extraction in a manner that will make the content extraction process automatic. Word spotting is a particular application of automatic speech recognition techniques in which the vocabulary of interest is relatively ....

D. James, "The Application of Classical Information Retrieval Techniques to Spoken Documents," Ph.D. Dissertation, University of Cambridge, United Kingdom, 1995.

....the vocabulary of interest is relatively small. Vocabularies of concepts from the ontology, excepting NPC concepts, can be used in our case. It is the job of 5 the recognizer to pick out only occurrences of keywords from this vocabulary in the speech ( in our case audio object ) to be recognized [14]. The output of a wordspotter is typically a list of keyword hits in this audio object. For example, if the occurrence of the concept, NFL is determined in a particular audio object, the object s metadata is the concept, NFL. Fully automated content extraction may be employed [10] Manual ....

David James. The Application of Classical Information Retrieval Techniques to Spoken Documents. Ph.D. Thesis, University of Cambridge, United Kingdom, 1995.

....streams of data. Much research has been done on the problem of selecting relevant items from large collections of text documents given a request from a user [2] Only recently has there been work addressing the retrieval of information from other media such as images, video, audio, and speech [3, 5, 6, 9, 10]. Given the growing amounts of spoken language data, such as recorded speech messages and radio and television broadcasts, the development of automatic methods to index, organize, and retrieve spoken documents will become more important. In our previous work [6] we investigated the feasibility ....

....from the story headlines and are relatively short, each averaging 4.5 words. Each query has an average of 6.2 relevant documents. Although this data set is small in comparison to experimental text retrieval collections [2] it is comparable to data sets used in other speech retrieval experiments [3, 5, 9]. 6. EXPERIMENTS AND RESULTS 6.1. Phonetic Recognition Experiments A series of phonetic recognition experiments is performed exploring the effects of using different acoustic and language models to try to improve phonetic recognition performance. 6.1.1. Segment Acoustic Model The most basic ....

D.A. James, "The Application of Classical Information Retrieval Techniques to Spoken Documents," Ph.D. Thesis, University of Cambridge, UK, 1995.

....More recently, ASR research has broadened its scope to include the transcription of general audio data (GAD) from sources such as radio, television, or movies. This shift in research focus is largely brought on by the growing need to shift content based information retrieval from text to speech [5], so that the computer can satisfy requests such as, Play me the speech by President Kennedy in which he said, Ich bin ein Berliner. GAD pose new challenges to present day ASR technology because they often contain extemporaneously generated, and therefore disfluent speech, with words drawn ....

James, David Anthony. The Application of Classical Information Retrieval Techniques to Spoken Documents." PhD thesis, Univ. of Cambridge, Feb., 1995.

....have created a completely new information era. Intelligent and efficient information retrieval techniques providing Internet users with easy access to spoken documents, such as broadcast radio and television programs, become highly desired and have been extensively studied in recent years [1 6]. At the same time, the DARPA Hub 4 contest that began in 1995 has been evaluating the technologies of use of large vocabulary continuous speech recognition (LVCSR) to transcribe audio recordings of broadcast news with many evaluation results reported [7] so far. Regardless of all these ....

D. A. James, "The Application of Classical Information Retrieval to Techniques to Spoken Documents," Ph.D. thesis, University of Cambridge, UK, 1995.

.... of document and text retrieval, on the problem of selecting relevant items from a large collection of text documents given a request from a user [18,37,40] Only recently has there been work addressing the retrieval of information from other media such as images, audio, video, and speech [9,12,21,23,49,51]. This expansion into other media raises new and interesting research issues in information retrieval especially in the areas of content representation and performance robustness. Given that increasingly large portions of the available data contain spoken language information, such as recorded ....

.... use phone sequences (phone n grams) generated by post processing the output of a phonetic speech recognizer [31,49] There are also methods that search for the query terms on phonetic transcriptions or phone lattice representations of the speech messages instead of creating subword indexing terms [9,23,26,48]. Some of these methods combine subword and large vocabulary word approaches to address the issue of new words in the queries [23,26] However, there have been few studies that explore the space of possible subword unit representations to determine the complexity of the subword units needed to ....

[Article contains additional citation context not shown here]

D. A. James, The Application of Classical Information Retrieval Techniques to Spoken Documents. Ph.D. thesis, University of Cambridge, Cambridge, U.K., 1995.

....posterior probability estimates derived from the recurrent network acoustic model, in order to detect index terms not in the recognizer s vocabulary. Jones et al. 1996) used a continuousspeech large vocabulary recognition system in combination with the phone lattice based word spotting method of James (1995). They showed that the two methods are complementary and work best in combination. In phoneme based solutions, one possibility is to expand the query with errorful variants of the original terms, in order to improve the chance of matching wrongly recognized terms. A similar method is to expand the ....

James, D. A. 1995. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD Thesis, Engineering Dept., University of Cambridge, England.

....perfect phone transcription: the best automatic system to date is little better than 70 accurate [Robinson et al. 1994] which severely limits the ability to accurately find arbitrary phone sequences since even one phone recognition error will result in a search miss. Another method developed by [James, 1995] uses a more sophisticated recogniser to generate a phone lattice that contains multiple phone hypotheses. Thus when the lattice is searched for query terms several different phones will be considered at each point. Often many putative term occurrences will be found, but unfortunately a large ....

....speech data is less than ours. There are also fewer requests. This work has explored various alternative methods of handling phone sequences. However, these tests and ours are so different in the data used, as well as the speech processing methods, that a straightforward comparison is impossible. [James, 1995] reports experiments using all his test collection query terms as keywords along with a phone lattice. His results, for a news data set somewhat smaller than ours, show about the same relative performance for spoken document retrieval against text transcriptions as ours. He has also, like us, ....

James, D. A. (1995). The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University.

.... and in particular there exists a large amount of training data [Lamel et al. 1986] Garofolo et al. 1990] LDC, 1992] At Cambridge University, a video mail retrieval system is being developed that currently accepts 35 query words [Sparck Jones et al. 1995] Similar work was also done by David James [1995] who experimented on retrieval of English news broadcasts. At Carnegie Mellon University, a retrieval system is being developed for a digital video library [Hauptmann et al. 1995] Informedia, 1996] As our main contribution, we present a family of speech retrieval methods suitable for a ....

James, D. (1995). The Application of Classical Information Retrieval to Techniques to Spoken Documents. PhD thesis, University of Cambridge, UK.

....used the output of a speech recognizer to index speech using standard IR techniques. To attack the out of vocabulary problem and the misrecognition problem, on the fly word spotting was done for the query words [16] using a fast phone lattice based word spotting algorithm developed by James [11]. Document scores from the recognizer s transcript and from the word spotting results were combined to get a final document score. Another approach to speech retrieval is to recognize subword units (for instance, phones) and use sequences of these sub word units as index terms. These approaches ....

D.A. James. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University, England, 1995.

.... More recently, ASR research has broadened its scope to include the transcription of general audio data (GAD) from sources such as radio or television [2] This shift in research focus is largely brought on by the growing need to shift contentbased information retrieval from text to speech [6], so that the computer can satisfy requests such as, Play me the speech by President Kennedy in which he said, Ich bin ein Berliner. GAD pose new challenges to present day ASR technology because they often contain extemporaneously generated, and therefore disfluent speech, with words drawn ....

James, David Anthony. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Univ. of Cambridge, February 1995.

.... amount of oversight over the surveillance process being allowed [FIfF 1995] Given that many intelligence agencies already have the capability to scan voice communications for individual voices and keywords (for example [CSE 1993] using technology which is easily available (see for example [James 1995], which covers speech recognition and automatic topic classification with scanning for items matching an arbitrary expression of the information requirement) and that a recent change to the German G10 law specifically allows for this form of scanning, there is potential for largescale automated ....

James, D. "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD thesis, Downing College, UK, 1995.

No context found.

James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, Speech, Vision and Robotic Group, Cambridge, U.K., February 1995.

No context found.

James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, February 1995.

No context found.

James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, February 1995.

No context found.

David James, The application of Classical Information Retrieval Techniques to Spoken Documents, Thesis, University of Cambridge, 1995.

No context found.

James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, Speech, Vision and Robotic Group, Cambridge, U.K., February 1995.

No context found.

James D. A., "The Application of Classical Information Retrieval Techniques to Spoken Documents", PhD Thesis, University of Cambridge, February 1995.

No context found.

D. A. James. The Application of Classical Information Retrieval Techniques to Spoken Documents. PhD thesis, Cambridge University,February 1995.

No context found.

-236. James, D. A. (1995). The Application of Classical Information Retrieval Techniques to Spoken Documents. Ph. D. thesis, University of Cambridge, Cambridge, U.K.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC