Pub Date : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-60
Eszter Iklódi, Gábor Recski, Gábor Borbély, María José Castro Bleda
This paper proposes a novel method for finding linear mappings among word vectors for various languages. Compared to previous approaches, this method does not learn translation matrices between two specific languages, but between a given language and a shared, universal space. The system was trained in two different modes, first between two languages, and after that applying three languages at the same time. In the first case two different training data were applied; Dinu’s English-Italian benchmark data [1], and English-Italian translation pairs extracted from the PanLex database [2]. In the second case only the PanLex database was used. The system performs on English-Italian languages with the best setting significantly better than the baseline system of Mikolov et al. [3], and it provides a comparable performance with the more sophisticated systems of Faruqui and Dyer [4] and Dinu et al. [1]. Exploiting the richness of the PanLex database, the proposed method makes it possible to learn linear mappings among an arbitrary number languages.
{"title":"Building a global dictionary for semantic technologies","authors":"Eszter Iklódi, Gábor Recski, Gábor Borbély, María José Castro Bleda","doi":"10.21437/IBERSPEECH.2018-60","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-60","url":null,"abstract":"This paper proposes a novel method for finding linear mappings among word vectors for various languages. Compared to previous approaches, this method does not learn translation matrices between two specific languages, but between a given language and a shared, universal space. The system was trained in two different modes, first between two languages, and after that applying three languages at the same time. In the first case two different training data were applied; Dinu’s English-Italian benchmark data [1], and English-Italian translation pairs extracted from the PanLex database [2]. In the second case only the PanLex database was used. The system performs on English-Italian languages with the best setting significantly better than the baseline system of Mikolov et al. [3], and it provides a comparable performance with the more sophisticated systems of Faruqui and Dyer [4] and Dinu et al. [1]. Exploiting the richness of the PanLex database, the proposed method makes it possible to learn linear mappings among an arbitrary number languages.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"1 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":"130134261","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.2018-39
Benjamin Maurice, H. Bredin, Ruiqing Yin, Jose Patino, H. Delgado, C. Barras, N. Evans, Camille Guinaudeau
This paper describes ODESSA and PLUMCOT submissions to Albayzin Multimodal Diarization Challenge 2018. Given a list of people to recognize (alongside image and short video samples of those people), the task consists in jointly answering the two questions “who speaks when?” and “who appears when?”. Both consortia submitted 3 runs (1 primary and 2 contrastive) based on the same underlying mono-modal neural technologies : neural speaker segmentation, neural speaker embeddings, neural face embeddings, and neural talking-face detection. Our submissions aim at showing that face clustering and recognition can (hopefully) help to improve speaker diarization.
{"title":"ODESSA/PLUMCOT at Albayzin Multimodal Diarization Challenge 2018","authors":"Benjamin Maurice, H. Bredin, Ruiqing Yin, Jose Patino, H. Delgado, C. Barras, N. Evans, Camille Guinaudeau","doi":"10.21437/IBERSPEECH.2018-39","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-39","url":null,"abstract":"This paper describes ODESSA and PLUMCOT submissions to Albayzin Multimodal Diarization Challenge 2018. Given a list of people to recognize (alongside image and short video samples of those people), the task consists in jointly answering the two questions “who speaks when?” and “who appears when?”. Both consortia submitted 3 runs (1 primary and 2 contrastive) based on the same underlying mono-modal neural technologies : neural speaker segmentation, neural speaker embeddings, neural face embeddings, and neural talking-face detection. Our submissions aim at showing that face clustering and recognition can (hopefully) help to improve speaker diarization.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"14 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":"130906457","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.2018-16
D. T. Santiago, Ian Benderitter, C. García-Mateo
Automatic sign language recognition (ASLR) is quite a complex task, not only for the intrinsic difficulty of automatic video information retrieval, but also because almost every sign language (SL) can be considered as an under-resourced language when it comes to language technology. Spanish sign language (SSL) is one of those under-resourced languages. Developing technology for SSL implies a number of technical challenges that must be tackled down in a structured and sequential manner. In this paper, the problem of how to design an experimental framework for machine-learning-based ASLR is addressed. In our review of existing datasets, our main conclusion is that there is a need for high-quality data. We therefore propose some guidelines on how to conduct the acquisition and annotation of an SSL dataset. These guidelines were developed after conducting some preliminary ASLR experiments with small and limited subsets of existing datasets.
{"title":"Experimental Framework Design for Sign Language Automatic Recognition","authors":"D. T. Santiago, Ian Benderitter, C. García-Mateo","doi":"10.21437/iberspeech.2018-16","DOIUrl":"https://doi.org/10.21437/iberspeech.2018-16","url":null,"abstract":"Automatic sign language recognition (ASLR) is quite a complex task, not only for the intrinsic difficulty of automatic video information retrieval, but also because almost every sign language (SL) can be considered as an under-resourced language when it comes to language technology. Spanish sign language (SSL) is one of those under-resourced languages. Developing technology for SSL implies a number of technical challenges that must be tackled down in a structured and sequential manner. In this paper, the problem of how to design an experimental framework for machine-learning-based ASLR is addressed. In our review of existing datasets, our main conclusion is that there is a need for high-quality data. We therefore propose some guidelines on how to conduct the acquisition and annotation of an SSL dataset. These guidelines were developed after conducting some preliminary ASLR experiments with small and limited subsets of existing datasets.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"34 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":"128858848","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.2018-17
Cassio T. Batista, Ana Larissa Dias, N. Neto
Kaldi has become a very popular toolkit for automatic speech recognition, showing considerable improvements through the combination of hidden Markov models (HMM) and deep neural networks (DNN). However, in spite of its great performance for some languages (e.g. English, Italian, Serbian, etc.), the resources for Brazilian Portuguese (BP) are still quite limited. This work describes what appears to be the first attempt to cre-ate Kaldi-based scripts and baseline acoustic models for BP using Kaldi tools. Experiments were carried out for dictation tasks and a comparison to CMU Sphinx toolkit in terms of word error rate (WER) was performed. Results seem promising, since Kaldi achieved the absolute lowest WER of 4.75% with HMM-DNN and outperformed CMU Sphinx even when using Gaussian mixture models only.
{"title":"Baseline Acoustic Models for Brazilian Portuguese Using Kaldi Tools","authors":"Cassio T. Batista, Ana Larissa Dias, N. Neto","doi":"10.21437/IberSPEECH.2018-17","DOIUrl":"https://doi.org/10.21437/IberSPEECH.2018-17","url":null,"abstract":"Kaldi has become a very popular toolkit for automatic speech recognition, showing considerable improvements through the combination of hidden Markov models (HMM) and deep neural networks (DNN). However, in spite of its great performance for some languages (e.g. English, Italian, Serbian, etc.), the resources for Brazilian Portuguese (BP) are still quite limited. This work describes what appears to be the first attempt to cre-ate Kaldi-based scripts and baseline acoustic models for BP using Kaldi tools. Experiments were carried out for dictation tasks and a comparison to CMU Sphinx toolkit in terms of word error rate (WER) was performed. Results seem promising, since Kaldi achieved the absolute lowest WER of 4.75% with HMM-DNN and outperformed CMU Sphinx even when using Gaussian mixture models only.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"112 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":"116023191","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}
Pub Date : 2018-11-21DOI: 10.21437/IBERSPEECH.2018-57
Nazim Dugan, C. Glackin, Gérard Chollet, Nigel Cannings
This paper describes the system developed by the Empathic team for the open set condition of the Iberspeech 2018 Speech to Text Transcription Challenge. A DNN-HMM hybrid acoustic model is developed, with MFCC's and iVectors as input features, using the Kaldi framework. The provided ground truth transcriptions for training and development are cleaned up using customized clean-up scripts and then realigned using a two-step alignment procedure which uses word lattice results coming from a previous ASR system. 261 hours of data is selected from train and dev1 subsections of the provided data, by applying a selection criterion on the utterance level scoring results. The selected data is merged with the 91 hours of training data used to train the previous ASR system with a factor 3 times data augmentation by reverberation using a noise corpus on the total training data, resulting a total of 1057 hours of final …
本文描述了移情团队为Iberspeech 2018 Speech to Text Transcription Challenge的开放设置条件开发的系统。采用Kaldi框架，建立了以MFCC和矢量为输入特征的DNN-HMM混合声学模型。为培训和发展提供的基础真相转录使用定制的清理脚本进行清理，然后使用两步对齐程序重新对齐，该程序使用来自先前ASR系统的词格结果。通过对话语水平评分结果应用选择标准，从所提供数据的train和dev1小节中选择261小时的数据。选择的数据与之前用于训练ASR系统的91小时训练数据合并，并在总训练数据上使用噪声语料库进行混响，使数据增加3倍，从而获得总计1057小时的最终数据。
{"title":"Intelligent Voice ASR system for Iberspeech 2018 Speech to Text Transcription Challenge","authors":"Nazim Dugan, C. Glackin, Gérard Chollet, Nigel Cannings","doi":"10.21437/IBERSPEECH.2018-57","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-57","url":null,"abstract":"This paper describes the system developed by the Empathic team for the open set condition of the Iberspeech 2018 Speech to Text Transcription Challenge. A DNN-HMM hybrid acoustic model is developed, with MFCC's and iVectors as input features, using the Kaldi framework. The provided ground truth transcriptions for training and development are cleaned up using customized clean-up scripts and then realigned using a two-step alignment procedure which uses word lattice results coming from a previous ASR system. 261 hours of data is selected from train and dev1 subsections of the provided data, by applying a selection criterion on the utterance level scoring results. The selected data is merged with the 91 hours of training data used to train the previous ASR system with a factor 3 times data augmentation by reverberation using a noise corpus on the total training data, resulting a total of 1057 hours of final …","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"218 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":"124297959","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: 10.4995/Thesis/10251/86137
Emilio Granell, C. Martínez-Hinarejos, Verónica Romero
Natural Language Processing (NLP) is an interdisciplinary research field of Computer Science, Linguistics, and Pattern Recognition that studies, among others, the use of human natural languages in Human-Computer Interaction (HCI). Most of NLP research tasks can be applied for solving real-world problems. This is the case of natural language recognition and natural language translation, that can be used for building automatic systems for document transcription and document translation. Regarding digitalised handwritten text documents, transcription is used to obtain an easy digital access to the contents, since simple image digitalisation only provides, in most cases, search by image and not by linguistic contents (keywords, expressions, syntactic or semantic categories). Transcription is even more important in historical manuscripts, since most of these documents are unique and the preservation of their contents is crucial for cultural and historical reasons. The transcription of historical manuscripts is usually done by paleographers, who are experts on ancient script and vocabulary. Recently, Handwritten Text Recognition (HTR) has become a common tool for assisting paleographers in their task, by providing a draft transcription that they may amend with more or less sophisticated methods. This draft transcription is useful when it presents an error rate low enough to make the amending process more comfortable than a complete transcription from scratch. Thus, obtaining a draft transcription with an acceptable low error rate is crucial to have this NLP technology incorporated into the transcription process. The work described in this thesis is focused on the improvement of the draft transcription offered by an HTR system, with the aim of reducing the effort made by paleographers for obtaining the actual transcription on digitalised historical manuscripts. This problem is faced from three different, but complementary, scenarios: · Multimodality: The use of HTR systems allow paleographers to speed up the manual transcription process, since they are able to correct on a draft transcription. Another alternative is to obtain the draft transcription by dictating the contents to an Automatic Speech Recognition (ASR) system. When both sources (image and speech) are available, a multimodal combination is possible and an iterative process can be used in order to refine the final hypothesis. · Interactivity: The use of assistive technologies in the transcription process allows one to reduce the time and human effort required for obtaining the actual transcription, given that the assistive system and the palaeographer cooperate to generate a perfect transcription. Multimodal feedback can be used to provide the assistive system with additional sources of information by using signals that represent the whole same sequence of words to transcribe (e.g. a text image, and the speech of the dictation of the contents of this text image), or that represent just
{"title":"Advances on the Transcription of Historical Manuscripts based on Multimodality, Interactivity and Crowdsourcing","authors":"Emilio Granell, C. Martínez-Hinarejos, Verónica Romero","doi":"10.4995/Thesis/10251/86137","DOIUrl":"https://doi.org/10.4995/Thesis/10251/86137","url":null,"abstract":"Natural Language Processing (NLP) is an interdisciplinary research field of Computer Science, Linguistics, and Pattern Recognition that studies, among others, the use of human natural languages in Human-Computer Interaction (HCI). Most of NLP research tasks can be applied for solving real-world problems. This is the case of natural language recognition and natural language translation, that can be used for building automatic systems for document transcription and document translation. \u0000Regarding digitalised handwritten text documents, transcription is used to obtain an easy digital access to the contents, since simple image digitalisation only provides, in most cases, search by image and not by linguistic contents (keywords, expressions, syntactic or semantic categories). Transcription is even more important in historical manuscripts, since most of these documents are unique and the preservation of their contents is crucial for cultural and historical reasons. \u0000The transcription of historical manuscripts is usually done by paleographers, who are experts on ancient script and vocabulary. Recently, Handwritten Text Recognition (HTR) has become a common tool for assisting paleographers in their task, by providing a draft transcription that they may amend with more or less sophisticated methods. This draft transcription is useful when it presents an error rate low enough to make the amending process more comfortable than a complete transcription from scratch. Thus, obtaining a draft transcription with an acceptable low error rate is crucial to have this NLP technology incorporated into the transcription process. \u0000The work described in this thesis is focused on the improvement of the draft transcription offered by an HTR system, with the aim of reducing the effort made by paleographers for obtaining the actual transcription on digitalised historical manuscripts. \u0000This problem is faced from three different, but complementary, scenarios: · Multimodality: The use of HTR systems allow paleographers to speed up the manual transcription process, since they are able to correct on a draft transcription. Another alternative is to obtain the draft transcription by dictating the contents to an Automatic Speech Recognition (ASR) system. When both sources (image and speech) are available, a multimodal combination is possible and an iterative process can be used in order to refine the final hypothesis. \u0000· Interactivity: The use of assistive technologies in the transcription process allows one to reduce the time and human effort required for obtaining the actual transcription, given that the assistive system and the palaeographer cooperate to generate a perfect transcription. \u0000Multimodal feedback can be used to provide the assistive system with additional sources of information by using signals that represent the whole same sequence of words to transcribe (e.g. a text image, and the speech of the dictation of the contents of this text image), or that represent just ","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"34 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":"121748460","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.2018-4
Esther Rituerto-González, A. Gallardo-Antolín, Carmen Peláez-Moreno
Speaker Recognition systems exhibit a decrease in performance when the input speech is not in optimal circumstances, for example when the user is under emotional or stress conditions. The objective of this paper is measuring the effects of stress on speech to ultimately try to mitigate its consequences on a speaker recognition task. On this paper, we develop a stress-robust speaker identification system using data selection and augmentation by means of the manipulation of the original speech utterances. An extensive experimentation has been carried out for assessing the effectiveness of the proposed techniques. First, we concluded that the best performance is always obtained when naturally stressed samples are included in the training set, and second, when these are not available, their substitution and augmentation with synthetically generated stress-like samples, improves the performance of the system.
{"title":"Speaker Recognition under Stress Conditions","authors":"Esther Rituerto-González, A. Gallardo-Antolín, Carmen Peláez-Moreno","doi":"10.21437/IBERSPEECH.2018-4","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-4","url":null,"abstract":"Speaker Recognition systems exhibit a decrease in performance when the input speech is not in optimal circumstances, for example when the user is under emotional or stress conditions. The objective of this paper is measuring the effects of stress on speech to ultimately try to mitigate its consequences on a speaker recognition task. On this paper, we develop a stress-robust speaker identification system using data selection and augmentation by means of the manipulation of the original speech utterances. An extensive experimentation has been carried out for assessing the effectiveness of the proposed techniques. First, we concluded that the best performance is always obtained when naturally stressed samples are included in the training set, and second, when these are not available, their substitution and augmentation with synthetically generated stress-like samples, improves the performance of the system.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"1 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":"130428050","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.2018-19
Pablo Gimeno, I. Viñals, A. Ortega, A. Miguel, EDUARDO LLEIDA SOLANO
This paper presents a new approach for automatic audio segmentation based on Recurrent Neural Networks. Our system takes advantage of the capability of Bidirectional Long Short Term Memory Networks (BLSTM) for modeling temporal dy-namics of the input signals. The DNN is complemented by a resegmentation module, gaining long-term stability by means of the tied-state concept in Hidden Markov Models. Further-more, feature exploration has been performed to best represent the information in the input data. The acoustic features that have been included are spectral log-filter-bank energies and musical features such as chroma. This new approach has been evaluated with the Albayz´ın 2010 audio segmentation evaluation dataset. The evaluation requires to differentiate five audio conditions: music, speech, speech with music, speech with noise and others. Competitive results were obtained, achieving a relative improvement of 15.75% compared to the best results found in the literature for this database.
{"title":"A Recurrent Neural Network Approach to Audio Segmentation for Broadcast Domain Data","authors":"Pablo Gimeno, I. Viñals, A. Ortega, A. Miguel, EDUARDO LLEIDA SOLANO","doi":"10.21437/IBERSPEECH.2018-19","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-19","url":null,"abstract":"This paper presents a new approach for automatic audio segmentation based on Recurrent Neural Networks. Our system takes advantage of the capability of Bidirectional Long Short Term Memory Networks (BLSTM) for modeling temporal dy-namics of the input signals. The DNN is complemented by a resegmentation module, gaining long-term stability by means of the tied-state concept in Hidden Markov Models. Further-more, feature exploration has been performed to best represent the information in the input data. The acoustic features that have been included are spectral log-filter-bank energies and musical features such as chroma. This new approach has been evaluated with the Albayz´ın 2010 audio segmentation evaluation dataset. The evaluation requires to differentiate five audio conditions: music, speech, speech with music, speech with noise and others. Competitive results were obtained, achieving a relative improvement of 15.75% compared to the best results found in the literature for this database.","PeriodicalId":115963,"journal":{"name":"IberSPEECH Conference","volume":"53 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":"114068282","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.2018-48
Abbas Khosravani, C. Glackin, Nazim Dugan, G. Chollet, Nigel Cannings
This paper describes the Intelligent Voice (IV) speaker diarization system for IberSPEECH-RTVE 2018 speaker diarization challenge. We developed a new speaker diarization built on the success of deep neural network based speaker embeddings in speaker verification systems. In contrary to acoustic features such as MFCCs, deep neural network embeddings are much better at discerning speaker identities especially for speech acquired without constraint on recording equipment and environment. We perform spectral clustering on our proposed CNNLSTM-based speaker embeddings to find homogeneous segments and generate speaker log likelihood for each frame. A HMM is then used to refine the speaker posterior probabilities through limiting the probability of switching between speakers when changing frames. We present results obtained on the development set (dev2) as well as the evaluation set …
{"title":"The Intelligent Voice System for the IberSPEECH-RTVE 2018 Speaker Diarization Challenge","authors":"Abbas Khosravani, C. Glackin, Nazim Dugan, G. Chollet, Nigel Cannings","doi":"10.21437/IBERSPEECH.2018-48","DOIUrl":"https://doi.org/10.21437/IBERSPEECH.2018-48","url":null,"abstract":"This paper describes the Intelligent Voice (IV) speaker diarization system for IberSPEECH-RTVE 2018 speaker diarization challenge. We developed a new speaker diarization built on the success of deep neural network based speaker embeddings in speaker verification systems. In contrary to acoustic features such as MFCCs, deep neural network embeddings are much better at discerning speaker identities especially for speech acquired without constraint on recording equipment and environment. We perform spectral clustering on our proposed CNNLSTM-based speaker embeddings to find homogeneous segments and generate speaker log likelihood for each frame. A HMM is then used to refine the speaker posterior probabilities through limiting the probability of switching between speakers when changing frames. We present results obtained on the development set (dev2) as well as the evaluation set …","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":"114197936","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：