Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034639
H. Murao, Nobuo Kawaguchi, S. Matsubara, Y. Inagaki
This paper proposes a new query generation method that is based on examples of human-to-human dialogue. Along with modeling the information flow in dialogue, a system for information retrieval in-car has been designed. The system refers to the dialogue corpus to find an example that is similar to input speech, and makes a query from the example. We also give the experimental results to show the effectiveness of this method.
{"title":"Example-based query generation for spontaneous speech","authors":"H. Murao, Nobuo Kawaguchi, S. Matsubara, Y. Inagaki","doi":"10.1109/ASRU.2001.1034639","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034639","url":null,"abstract":"This paper proposes a new query generation method that is based on examples of human-to-human dialogue. Along with modeling the information flow in dialogue, a system for information retrieval in-car has been designed. The system refers to the dialogue corpus to find an example that is similar to input speech, and makes a query from the example. We also give the experimental results to show the effectiveness of this method.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"47 5-6","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114035589","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034621
T. Shinozaki, S. Furui
This paper proposes the use of decision trees for analyzing errors in spontaneous presentation speech recognition. The trees are designed to predict whether a word or a phoneme can be correctly recognized or not, using word or phoneme attributes as inputs. The trees, are constructed using training "cases" by choosing questions about attributes step by step according to the gain ratio criterion. The errors in recognizing spontaneous presentations given by 10 male speakers were analyzed, and the explanation capability of attributes for the recognition errors was quantitatively evaluated. A restricted set of attributes closely related to the recognition errors was identified for both words and phonemes.
{"title":"Error analysis using decision trees in spontaneous presentation speech recognition","authors":"T. Shinozaki, S. Furui","doi":"10.1109/ASRU.2001.1034621","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034621","url":null,"abstract":"This paper proposes the use of decision trees for analyzing errors in spontaneous presentation speech recognition. The trees are designed to predict whether a word or a phoneme can be correctly recognized or not, using word or phoneme attributes as inputs. The trees, are constructed using training \"cases\" by choosing questions about attributes step by step according to the gain ratio criterion. The errors in recognizing spontaneous presentations given by 10 male speakers were analyzed, and the explanation capability of attributes for the recognition errors was quantitatively evaluated. A restricted set of attributes closely related to the recognition errors was identified for both words and phonemes.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122702778","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034646
Alexander Hauptmann, Wei-Hao Lin
The Informedia Project digital video library pioneered the automatic analysis of television broadcast news and its retrieval on demand. Building on that system, we have developed a wearable, personalized Informedia system, which listens to and transcribes the wearer's part of a conversation, recognizes the face of the current dialog partner and remembers his/her voice. The next time the system sees the same person's face and hears the same voice, it can retrieve the audio from the last conversation, replaying in compressed form the names and major issues that were mentioned. All of this happens unobtrusively, somewhat like an intelligent assistant who whispers to you: "That's Bob Jones from Tech Solutions; two weeks ago in London you discussed solar panels". This paper outlines the general system components as well as interface considerations. Initial implementations showed that both face recognition methods and speaker identification technology have serious shortfalls that must be overcome.
{"title":"Beyond the Informedia digital video library: video and audio analysis for remembering conversations","authors":"Alexander Hauptmann, Wei-Hao Lin","doi":"10.1109/ASRU.2001.1034646","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034646","url":null,"abstract":"The Informedia Project digital video library pioneered the automatic analysis of television broadcast news and its retrieval on demand. Building on that system, we have developed a wearable, personalized Informedia system, which listens to and transcribes the wearer's part of a conversation, recognizes the face of the current dialog partner and remembers his/her voice. The next time the system sees the same person's face and hears the same voice, it can retrieve the audio from the last conversation, replaying in compressed form the names and major issues that were mentioned. All of this happens unobtrusively, somewhat like an intelligent assistant who whispers to you: \"That's Bob Jones from Tech Solutions; two weeks ago in London you discussed solar panels\". This paper outlines the general system components as well as interface considerations. Initial implementations showed that both face recognition methods and speaker identification technology have serious shortfalls that must be overcome.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"102 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131517073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034681
G. Hernández-Abrego, X. Menéndez-Pidal, L. Olorenshaw
A new confidence measure for isolated command recognition is presented. It is versatile and efficient in two ways. First, it is based exclusively on the speech recognizer's output. In addition, it is robust to changes in the vocabulary, acoustic model and parameter settings. Its calculation is very simple and it is based on the computation of a pseudo-filler score from an N-best list. Performance is tested in two different command recognition applications. Finally, it is efficient to separate correct results both from incorrect ones and from false alarms caused by out-of-vocabulary elements and noise.
{"title":"Robust and efficient confidence measure for isolated command recognition","authors":"G. Hernández-Abrego, X. Menéndez-Pidal, L. Olorenshaw","doi":"10.1109/ASRU.2001.1034681","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034681","url":null,"abstract":"A new confidence measure for isolated command recognition is presented. It is versatile and efficient in two ways. First, it is based exclusively on the speech recognizer's output. In addition, it is robust to changes in the vocabulary, acoustic model and parameter settings. Its calculation is very simple and it is based on the computation of a pseudo-filler score from an N-best list. Performance is tested in two different command recognition applications. Finally, it is efficient to separate correct results both from incorrect ones and from false alarms caused by out-of-vocabulary elements and noise.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133805377","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034608
J. Stadermann, G. Rigoll
Distributed speech recognition (DSR) is an interesting technology for mobile recognition tasks where the recognizer is split up into two parts and connected by a transmission channel. We compare the performance of standard and hybrid modeling approaches in this environment. The evaluation is done on clean and noisy speech samples taken from the TI digits and the Aurora databases. Our results show that, for this task, the hybrid modeling techniques can outperform standard continuous systems.
{"title":"Comparison of standard and hybrid modeling techniques for distributed speech recognition","authors":"J. Stadermann, G. Rigoll","doi":"10.1109/ASRU.2001.1034608","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034608","url":null,"abstract":"Distributed speech recognition (DSR) is an interesting technology for mobile recognition tasks where the recognizer is split up into two parts and connected by a transmission channel. We compare the performance of standard and hybrid modeling approaches in this environment. The evaluation is done on clean and noisy speech samples taken from the TI digits and the Aurora databases. Our results show that, for this task, the hybrid modeling techniques can outperform standard continuous systems.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"127 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133857897","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034684
H. Nakano
Popular Japanese mobile Web-phones are widely used to connect to Internet providers (IP). The most popular service on mobile Web-phones is E-mail. Currently, users type the messages using the ten standard keys on the phone. Several letters and Kana (Japanese phonetic characters) are assigned to each key, and the user steps through them by tapping the key repeatedly. After inputting several words, the user converts them into Kanji (Chinese character). Kana-Kanji conversion is still improving, and recently fast text input methods have been introduced, but these key input methods are still troublesome. A speech interface is expected to overcome this input difficulty. However, speech interfaces suffer several problems, both technical and social. The paper summarises these problems and looks at some methods by which technical solutions may be found.
{"title":"Introduction of speech interface for mobile information services","authors":"H. Nakano","doi":"10.1109/ASRU.2001.1034684","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034684","url":null,"abstract":"Popular Japanese mobile Web-phones are widely used to connect to Internet providers (IP). The most popular service on mobile Web-phones is E-mail. Currently, users type the messages using the ten standard keys on the phone. Several letters and Kana (Japanese phonetic characters) are assigned to each key, and the user steps through them by tapping the key repeatedly. After inputting several words, the user converts them into Kanji (Chinese character). Kana-Kanji conversion is still improving, and recently fast text input methods have been introduced, but these key input methods are still troublesome. A speech interface is expected to overcome this input difficulty. However, speech interfaces suffer several problems, both technical and social. The paper summarises these problems and looks at some methods by which technical solutions may be found.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"161 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115548766","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034581
F. Thiele, R. Bippus
Many techniques for speaker adaptation have been successfully applied to automatic speech recognition. This paper compares the performance of several adaptation methods with respect to their memory need and processing demand. For adaptation of a compact acoustic model with 4k densities, eigenvoices and structural MAP (SMAP) are investigated next to the well-known techniques of MAP (maximum a posteriori) and MLLR (maximum likelihood linear regression) adaptation. Experimental results are reported for unsupervised on-line adaptation on different amounts of adaptation data ranging from 4 to 500 words per speaker. The results show that for small amounts of adaptation data it might be more efficient to employ a larger baseline acoustic model without adaptation. Eigenvoices achieve the lowest word error rates of all adaptation techniques but SMAP presents a good compromise between memory requirement and accuracy.
{"title":"A comparative study of model-based adaptation techniques for a compact speech recognizer","authors":"F. Thiele, R. Bippus","doi":"10.1109/ASRU.2001.1034581","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034581","url":null,"abstract":"Many techniques for speaker adaptation have been successfully applied to automatic speech recognition. This paper compares the performance of several adaptation methods with respect to their memory need and processing demand. For adaptation of a compact acoustic model with 4k densities, eigenvoices and structural MAP (SMAP) are investigated next to the well-known techniques of MAP (maximum a posteriori) and MLLR (maximum likelihood linear regression) adaptation. Experimental results are reported for unsupervised on-line adaptation on different amounts of adaptation data ranging from 4 to 500 words per speaker. The results show that for small amounts of adaptation data it might be more efficient to employ a larger baseline acoustic model without adaptation. Eigenvoices achieve the lowest word error rates of all adaptation techniques but SMAP presents a good compromise between memory requirement and accuracy.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114219640","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034648
F. Wessel, H. Ney
For speech recognition systems, the amount of acoustic training data is of crucial importance. In the past, large amounts of speech were recorded and transcribed manually for training. Since untranscribed speech is available in various forms these days, the unsupervised training of a speech recognizer on recognized transcriptions is studied. A low-cost recognizer trained with only one hour of manually transcribed speech is used to recognize 72 hours of untranscribed acoustic data. These transcriptions are then used in combination with confidence measures to train an improved recognizer. The effect of confidence measures which are used to detect possible recognition errors is studied systematically. Finally, the unsupervised training is applied iteratively. Using this method, the recognizer is trained with very little manual effort while losing only 14.3% relative on the Broadcast News '96 and 18.6% relative on the Broadcast News '98 evaluation test sets.
{"title":"Unsupervised training of acoustic models for large vocabulary continuous speech recognition","authors":"F. Wessel, H. Ney","doi":"10.1109/ASRU.2001.1034648","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034648","url":null,"abstract":"For speech recognition systems, the amount of acoustic training data is of crucial importance. In the past, large amounts of speech were recorded and transcribed manually for training. Since untranscribed speech is available in various forms these days, the unsupervised training of a speech recognizer on recognized transcriptions is studied. A low-cost recognizer trained with only one hour of manually transcribed speech is used to recognize 72 hours of untranscribed acoustic data. These transcriptions are then used in combination with confidence measures to train an improved recognizer. The effect of confidence measures which are used to detect possible recognition errors is studied systematically. Finally, the unsupervised training is applied iteratively. Using this method, the recognizer is trained with very little manual effort while losing only 14.3% relative on the Broadcast News '96 and 18.6% relative on the Broadcast News '98 evaluation test sets.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115408833","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034611
T. Krisjansson, B. Frey, L. Deng, A. Acero
The performance of speech cleaning and noise adaptation algorithms is heavily dependent on the quality of the noise and channel models. Various strategies have been proposed in the literature for adapting to the current noise and channel conditions. We describe the joint learning of noise and channel distortion in a novel framework called ALGONQUIN. The learning algorithm employs a generalized EM strategy wherein the E step is approximate. We discuss the characteristics of the new algorithm, with a focus on convergence rates and parameter initialization. We show that the learning algorithm can successfully disentangle the non-linear effects of noise and linear effects of the channel and achieve a relative reduction in WER of 21.8% over the non-adaptive algorithm.
{"title":"Joint estimation of noise and channel distortion in a generalized EM framework","authors":"T. Krisjansson, B. Frey, L. Deng, A. Acero","doi":"10.1109/ASRU.2001.1034611","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034611","url":null,"abstract":"The performance of speech cleaning and noise adaptation algorithms is heavily dependent on the quality of the noise and channel models. Various strategies have been proposed in the literature for adapting to the current noise and channel conditions. We describe the joint learning of noise and channel distortion in a novel framework called ALGONQUIN. The learning algorithm employs a generalized EM strategy wherein the E step is approximate. We discuss the characteristics of the new algorithm, with a focus on convergence rates and parameter initialization. We show that the learning algorithm can successfully disentangle the non-linear effects of noise and linear effects of the channel and achieve a relative reduction in WER of 21.8% over the non-adaptive algorithm.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"452 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126288159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2001-12-09DOI: 10.1109/ASRU.2001.1034619
M. Novak, R. Mammone
This paper describes the use of exponential models to improve non-negative matrix factorization (NMF) based topic language models for automatic speech recognition. This modeling technique borrows the basic idea from latent semantic analysis (LSA), which is typically used in information retrieval. An improvement was achieved when exponential models were used to estimate the a posteriori topic probabilities for an observed history. This method improved the perplexity of the NMF model, resulting in a 24% perplexity improvement overall when compared to a trigram language model.
{"title":"Improvement of non-negative matrix factorization based language model using exponential models","authors":"M. Novak, R. Mammone","doi":"10.1109/ASRU.2001.1034619","DOIUrl":"https://doi.org/10.1109/ASRU.2001.1034619","url":null,"abstract":"This paper describes the use of exponential models to improve non-negative matrix factorization (NMF) based topic language models for automatic speech recognition. This modeling technique borrows the basic idea from latent semantic analysis (LSA), which is typically used in information retrieval. An improvement was achieved when exponential models were used to estimate the a posteriori topic probabilities for an observed history. This method improved the perplexity of the NMF model, resulting in a 24% perplexity improvement overall when compared to a trigram language model.","PeriodicalId":118671,"journal":{"name":"IEEE Workshop on Automatic Speech Recognition and Understanding, 2001. ASRU '01.","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2001-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124884836","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：