Pub Date : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-36
Alicia Lozano-Diez, J. González-Rodríguez, J. Gonzalez-Dominguez
In this manuscript, we summarize the findings presented in Alicia Lozano Diez’s Ph.D. Thesis, defended on the 22nd of June, 2018 in Universidad Autonoma de Madrid (Spain). In particular, this Ph.D. Thesis explores different approaches to the tasks of language and speaker recognition, focusing on systems where deep neural networks (DNNs) become part of traditional pipelines, replacing some stages or the whole system itself. First, we present a DNN as classifier for the task of language recognition. Second, we analyze the use of DNNs for feature extraction at frame-level, the so-called bottleneck features, for both language and speaker recognition. Finally, utterance-level representation of the speech segments learned by the DNN (known as embedding) is described and presented for the task of language recognition. All these approaches provide alter-natives to classical language and speaker recognition systems based on i-vectors (Total Variability modeling) over acoustic features (MFCCs, for instance). Moreover, they usually yield better results in terms of performance. stochastic gradient descent to minimize the negative log-likelihood. We conducted experiments to evaluate the influence of differ-IberSPEECH
{"title":"Bottleneck and Embedding Representation of Speech for DNN-based Language and Speaker Recognition","authors":"Alicia Lozano-Diez, J. González-Rodríguez, J. Gonzalez-Dominguez","doi":"10.21437/IBERSPEECH.2018-36","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-36","url":null,"abstract":"In this manuscript, we summarize the findings presented in Alicia Lozano Diez’s Ph.D. Thesis, defended on the 22nd of June, 2018 in Universidad Autonoma de Madrid (Spain). In particular, this Ph.D. Thesis explores different approaches to the tasks of language and speaker recognition, focusing on systems where deep neural networks (DNNs) become part of traditional pipelines, replacing some stages or the whole system itself. First, we present a DNN as classifier for the task of language recognition. Second, we analyze the use of DNNs for feature extraction at frame-level, the so-called bottleneck features, for both language and speaker recognition. Finally, utterance-level representation of the speech segments learned by the DNN (known as embedding) is described and presented for the task of language recognition. All these approaches provide alter-natives to classical language and speaker recognition systems based on i-vectors (Total Variability modeling) over acoustic features (MFCCs, for instance). Moreover, they usually yield better results in terms of performance. stochastic gradient descent to minimize the negative log-likelihood. We conducted experiments to evaluate the influence of differ-IberSPEECH","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133548364","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 : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-61
J. Garrido, Marta Codina, K. Fodge
This paper presents TransDic, a free distribution tool for the phonetic transcription of word lists in Spanish and Catalan which allows the generation of phonetic transcription variants, a feature that can be useful for some technological applications, such as speech recognition. It allows the transcription in both standard Spanish and Catalan, but also in several dialects of these two languages spoken in Spain. Its general structure, input, output and main functionalities are presented, and the procedure followed to define and implement the transcription rules in the tool is described. Finally, the results of an evaluation carried for both languages are presented, which show that TransDic performs correctly the transcription tasks that it was developed for.
{"title":"TransDic, a public domain tool for the generation of phonetic dictionaries in standard and dialectal Spanish and Catalan","authors":"J. Garrido, Marta Codina, K. Fodge","doi":"10.21437/IBERSPEECH.2018-61","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-61","url":null,"abstract":"This paper presents TransDic, a free distribution tool for the phonetic transcription of word lists in Spanish and Catalan which allows the generation of phonetic transcription variants, a feature that can be useful for some technological applications, such as speech recognition. It allows the transcription in both standard Spanish and Catalan, but also in several dialects of these two languages spoken in Spain. Its general structure, input, output and main functionalities are presented, and the procedure followed to define and implement the transcription rules in the tool is described. Finally, the results of an evaluation carried for both languages are presented, which show that TransDic performs correctly the transcription tasks that it was developed for.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127408045","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 : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-51
Maria Cabello, D. Toledano, Javier Tejedor
Query-by-Example Spoken Term Detection is the task of detecting query occurrences within speech data (henceforth utterances). Our submission is based on a language-independent template matching approach. First, queries and utterances are represented as phonetic posteriorgrams computed for English language with the phoneme decoder developed by the Brno Uni-versity of Technology. Next, the Subsequence Dynamic Time Warping algorithm with a modified Pearson correlation coefficient as cost measure is employed to hipothesize detections. Results on development data showed an ATWV=0.1774 with MAVIR data and an ATWV=0.0365 with RTVE data.
{"title":"AUDIAS-CEU: A Language-independent approach for the Query-by-Example Spoken Term Detection task of the Search on Speech ALBAYZIN 2018 evaluation","authors":"Maria Cabello, D. Toledano, Javier Tejedor","doi":"10.21437/IBERSPEECH.2018-51","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-51","url":null,"abstract":"Query-by-Example Spoken Term Detection is the task of detecting query occurrences within speech data (henceforth utterances). Our submission is based on a language-independent template matching approach. First, queries and utterances are represented as phonetic posteriorgrams computed for English language with the phoneme decoder developed by the Brno Uni-versity of Technology. Next, the Subsequence Dynamic Time Warping algorithm with a modified Pearson correlation coefficient as cost measure is employed to hipothesize detections. Results on development data showed an ATWV=0.1774 with MAVIR data and an ATWV=0.0365 with RTVE data.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114513719","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 : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-13
Laura Cross Vila, Carlos Escolano, José A. R. Fonollosa, M. Costa-jussà
Speech Translation has been traditionally addressed with the concatenation of two tasks: Speech Recognition and Machine Translation. This approach has the main drawback that errors are concatenated. Recently, neural approaches to Speech Recognition and Machine Translation have made possible facing the task by means of an End-to-End Speech Translation architecture. In this paper, we propose to use the architecture of the Transformer which is based solely on attention-based mechanisms to address the End-to-End Speech Translation system. As a contrastive architecture, we use the same Transformer to built the Speech Recognition and Machine Translation systems to perform Speech Translation through concatenation of systems. Results on a Spanish-to-English standard task show that the end-to-end architecture is able to outperform the concatenated systems by half point BLEU.
{"title":"End-to-End Speech Translation with the Transformer","authors":"Laura Cross Vila, Carlos Escolano, José A. R. Fonollosa, M. Costa-jussà","doi":"10.21437/IBERSPEECH.2018-13","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-13","url":null,"abstract":"Speech Translation has been traditionally addressed with the concatenation of two tasks: Speech Recognition and Machine Translation. This approach has the main drawback that errors are concatenated. Recently, neural approaches to Speech Recognition and Machine Translation have made possible facing the task by means of an End-to-End Speech Translation architecture. In this paper, we propose to use the architecture of the Transformer which is based solely on attention-based mechanisms to address the End-to-End Speech Translation system. As a contrastive architecture, we use the same Transformer to built the Speech Recognition and Machine Translation systems to perform Speech Translation through concatenation of systems. Results on a Spanish-to-English standard task show that the end-to-end architecture is able to outperform the concatenated systems by half point BLEU.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131309104","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 : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-58
Laura Docío Fernández, C. García-Mateo
This paper describes the Speech-to-Text system developed by the Multimedia Technologies Group (GTM) of the atlanTTic research center at the University of Vigo, for the Albayzin Speech-to-Text Challenge (S2T) organized in the Iberspeech 2018 conference. The large vocabulary automatic speech recognition system is built using the Kaldi toolkit. It uses an hybrid Deep Neural Network - Hidden Markov Model (DNN-HMM) for acoustic modeling, and a rescoring of a trigram based word-lattices, obtained in a first decoding stage, with a fourgram language model or a language model based on a recurrent neural network. The system was evaluated only on the open set training condition.
{"title":"The GTM-UVIGO System for Albayzin 2018 Speech-to-Text Evaluation","authors":"Laura Docío Fernández, C. García-Mateo","doi":"10.21437/IBERSPEECH.2018-58","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-58","url":null,"abstract":"This paper describes the Speech-to-Text system developed by the Multimedia Technologies Group (GTM) of the atlanTTic research center at the University of Vigo, for the Albayzin Speech-to-Text Challenge (S2T) organized in the Iberspeech 2018 conference. The large vocabulary automatic speech recognition system is built using the Kaldi toolkit. It uses an hybrid Deep Neural Network - Hidden Markov Model (DNN-HMM) for acoustic modeling, and a rescoring of a trigram based word-lattices, obtained in a first decoding stage, with a fourgram language model or a language model based on a recurrent neural network. The system was evaluated only on the open set training condition.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"302 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122246221","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 : 2018-11-21DOI: 10.21437/IberSPEECH.2018-24
Mario Corrales-Astorgano, P. Martínez-Castilla, David Escudero Mancebo, L. Aguilar, César González Ferreras, Valentín Cardeñoso-Payo
{"title":"Towards an automatic evaluation of the prosody of people with Down syndrome","authors":"Mario Corrales-Astorgano, P. Martínez-Castilla, David Escudero Mancebo, L. Aguilar, César González Ferreras, Valentín Cardeñoso-Payo","doi":"10.21437/IberSPEECH.2018-24","DOIUrl":"https://doi.org/10.21437/IberSPEECH.2018-24","url":null,"abstract":"","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116689944","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 : 2018-11-21DOI: 10.21437/IberSPEECH.2018-11
I. Odriozola, I. Hernáez, E. Navas, Luis Serrano, Jon Sánchez
This work has been partially supported by the EU(FEDER) under grant TEC2015-67163-C2-1-R (RESTORE)(MINECO/FEDER, UE) and by the Basque Government undergrant KK-2017/00043 (BerbaOla)
{"title":"The observation likelihood of silence: analysis and prospects for VAD applications","authors":"I. Odriozola, I. Hernáez, E. Navas, Luis Serrano, Jon Sánchez","doi":"10.21437/IberSPEECH.2018-11","DOIUrl":"https://doi.org/10.21437/IberSPEECH.2018-11","url":null,"abstract":"This work has been partially supported by the EU(FEDER) under grant TEC2015-67163-C2-1-R (RESTORE)(MINECO/FEDER, UE) and by the Basque Government undergrant KK-2017/00043 (BerbaOla)","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132583163","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 : 2018-11-21DOI: 10.21437/IberSPEECH.2018-52
Luis Javier Rodriguez-Fuentes, M. Peñagarikano, A. Varona, Germán Bordel
This paper describes the systems developed by GTTS-EHU for the QbE-STD and STD tasks of the Albayzin 2018 Search on Speech Evaluation. Stacked bottleneck features (sBNF) are used as frame-level acoustic representation for both audio documents and spoken queries. In QbE-STD, a flavour of segmental DTW (originally developed for MediaEval 2013) is used to perform the search, which iteratively finds the match that minimizes the average distance between two test-normalized sBNF vectors, until either a maximum number of hits is obtained or the score does not attain a given threshold. The STD task is performed by synthesizing spoken queries (using publicly available TTS APIs), then averaging their sBNF representations and using the average query for QbE-STD. A publicly available toolkit (developed by BUT/Phonexia) has been used to extract three sBNF sets, trained for English monophone and triphone state posteriors (contrastive systems 3 and 4) and for multilingual triphone posteriors (contrastive system 2), respectively. The concatenation of the three sBNF sets has been also tested (contrastive system 1). The primary system consists of a discriminative fusion of the four contrastive systems. Detection scores are normalized on a query-by-query basis (qnorm), calibrated and, if two or more systems are considered, fused with other scores. Calibration and fusion parameters are discriminatively estimated using the ground truth of development data. Finally, due to a lack of robustness in calibration, Yes/No decisions are made by applying the MTWV thresholds obtained for the development sets, except for the COREMAH test set. In this case, calibration is based on the MAVIR corpus, and the 15% highest scores are taken as positive (Yes) detections.
{"title":"GTTS-EHU Systems for the Albayzin 2018 Search on Speech Evaluation","authors":"Luis Javier Rodriguez-Fuentes, M. Peñagarikano, A. Varona, Germán Bordel","doi":"10.21437/IberSPEECH.2018-52","DOIUrl":"https://doi.org/10.21437/IberSPEECH.2018-52","url":null,"abstract":"This paper describes the systems developed by GTTS-EHU for the QbE-STD and STD tasks of the Albayzin 2018 Search on Speech Evaluation. Stacked bottleneck features (sBNF) are used as frame-level acoustic representation for both audio documents and spoken queries. In QbE-STD, a flavour of segmental DTW (originally developed for MediaEval 2013) is used to perform the search, which iteratively finds the match that minimizes the average distance between two test-normalized sBNF vectors, until either a maximum number of hits is obtained or the score does not attain a given threshold. The STD task is performed by synthesizing spoken queries (using publicly available TTS APIs), then averaging their sBNF representations and using the average query for QbE-STD. A publicly available toolkit (developed by BUT/Phonexia) has been used to extract three sBNF sets, trained for English monophone and triphone state posteriors (contrastive systems 3 and 4) and for multilingual triphone posteriors (contrastive system 2), respectively. The concatenation of the three sBNF sets has been also tested (contrastive system 1). The primary system consists of a discriminative fusion of the four contrastive systems. Detection scores are normalized on a query-by-query basis (qnorm), calibrated and, if two or more systems are considered, fused with other scores. Calibration and fusion parameters are discriminatively estimated using the ground truth of development data. Finally, due to a lack of robustness in calibration, Yes/No decisions are made by applying the MTWV thresholds obtained for the development sets, except for the COREMAH test set. In this case, calibration is based on the MAVIR corpus, and the 15% highest scores are taken as positive (Yes) detections.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124213788","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 : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-49
Zili Huang, Leibny Paola García-Perera, J. Villalba, Daniel Povey, N. Dehak
We present the JHU system for Iberspeech-RTVE Speaker Diarization Evaluation. This assessment combines Spanish language and broadcast audio in the same recordings, conditions in which our system has not been tested before. To tackle this problem, the pipeline of our general system, developed en-tirely in Kaldi, includes an acoustic feature extraction, a SAD, an embedding extractor, a PLDA and a clustering stage. This pipeline was used for both, the open and the closed conditions (described in the evaluation plan). All the proposed solutions use wide-band data (16KHz) and MFCCs as their input. For the closed condition, the system trains a DNN SAD using the Albayzin2016 data. Due to the small amount of data available, the i-vector embedding extraction was the only approach explored for this task. The PLDA training utilizes Albayzin data fol-lowed by an Agglomerative Hierarchical Clustering (AHC) to obtain the speaker segmentation. The open condition employs the DNN SAD obtained in the closed condition. Four types of embeddings were extracted, x-vector-basic, x-vector-factored, i-vector-basic and BNF-i-vector. The x-vector-basic is a TDNN trained on augmented Voxceleb1 and Voxceleb2. The x-vector-factored is a factored-TDNN (TDNN-F) trained on SRE12-micphn, MX6-micphn, VoxCeleb and SITW-dev-core. The i-vector-basic was trained on Voxceleb1 and Voxceleb2 data (no augmentation). The BNF-i-vector is a BNF-posterior i-vector trained with the same data as x-vector-factored. The PLDA training for the new scenario uses the Albayzin2016 data. The four systems were fused at the score level. Once again, the AHC computed the final speaker segmentation. We tested our systems in the Albayzin2018 dev2 data and observed that the SAD is of importance to improve the results. Moreover, we noticed that x-vectors were better than i-vectors, as already observed in previous experiments.
{"title":"JHU Diarization System Description","authors":"Zili Huang, Leibny Paola García-Perera, J. Villalba, Daniel Povey, N. Dehak","doi":"10.21437/IBERSPEECH.2018-49","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-49","url":null,"abstract":"We present the JHU system for Iberspeech-RTVE Speaker Diarization Evaluation. This assessment combines Spanish language and broadcast audio in the same recordings, conditions in which our system has not been tested before. To tackle this problem, the pipeline of our general system, developed en-tirely in Kaldi, includes an acoustic feature extraction, a SAD, an embedding extractor, a PLDA and a clustering stage. This pipeline was used for both, the open and the closed conditions (described in the evaluation plan). All the proposed solutions use wide-band data (16KHz) and MFCCs as their input. For the closed condition, the system trains a DNN SAD using the Albayzin2016 data. Due to the small amount of data available, the i-vector embedding extraction was the only approach explored for this task. The PLDA training utilizes Albayzin data fol-lowed by an Agglomerative Hierarchical Clustering (AHC) to obtain the speaker segmentation. The open condition employs the DNN SAD obtained in the closed condition. Four types of embeddings were extracted, x-vector-basic, x-vector-factored, i-vector-basic and BNF-i-vector. The x-vector-basic is a TDNN trained on augmented Voxceleb1 and Voxceleb2. The x-vector-factored is a factored-TDNN (TDNN-F) trained on SRE12-micphn, MX6-micphn, VoxCeleb and SITW-dev-core. The i-vector-basic was trained on Voxceleb1 and Voxceleb2 data (no augmentation). The BNF-i-vector is a BNF-posterior i-vector trained with the same data as x-vector-factored. The PLDA training for the new scenario uses the Albayzin2016 data. The four systems were fused at the score level. Once again, the AHC computed the final speaker segmentation. We tested our systems in the Albayzin2018 dev2 data and observed that the SAD is of importance to improve the results. Moreover, we noticed that x-vectors were better than i-vectors, as already observed in previous experiments.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"697 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116117218","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：