{"id":451875,"date":"2017-12-21T19:30:05","date_gmt":"2017-12-22T03:30:05","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-event&#038;p=451875"},"modified":"2025-08-06T11:57:30","modified_gmt":"2025-08-06T18:57:30","slug":"interspeech-2018-special-session-low-resource-speech-recognition-challenge-indian-languages","status":"publish","type":"msr-event","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/event\/interspeech-2018-special-session-low-resource-speech-recognition-challenge-indian-languages\/","title":{"rendered":"Interspeech 2018 Special Session: Low Resource Speech Recognition Challenge for Indian Languages"},"content":{"rendered":"\n\n<p><strong>Organizing Committee:<\/strong><\/p>\n<ul>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/kalikab\/\">Kalika Bali<\/a><\/li>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/kdoss\/\">Krishna Doss Mohan<\/a><\/li>\n<li>Rupesh Kumar Mehta<\/li>\n<li>Niranjan Nayak<\/li>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/susitara\/\">Sunayana Sitaram<\/a><\/li>\n<li>Radhakrishnan Srikanth<\/li>\n<\/ul>\n<p><strong>Data preparation and baselines:<\/strong><\/p>\n<ul>\n<li>Brij Mohan Lal Srivastava<\/li>\n<li>Pallavi Matani<\/li>\n<li>Sandeepkumar Satpal<\/li>\n<li>Satarupa Guha<\/li>\n<li>Shambo Chatterjee<\/li>\n<li>Swapnajeet Padhi<\/li>\n<\/ul>\n<p><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p>In keeping with the Interspeech 2018 theme of \u2018Speech Research for Emerging Markets in Multilingual Societies\u2019, we\u00a0are organizing\u00a0a special session and challenge on speech recognition for low resource languages. Most languages in the world lack the amount of text, speech and linguistic resources required to build large Deep Neural Network (DNN)-based models. However, there have been many advances in DNN architectures, cross-lingual and multilingual speech processing techniques, and approaches incorporating linguistic knowledge into machine-learning based models, that can help in building systems for low resource languages. In this challenge, we\u00a0will focus on building Automatic Speech Recognition (ASR) systems for Indian languages with constraints on the data available for Acoustic Modeling and Language Modeling.<\/p>\n<p>India has around 1500 languages, of which 22 languages have been given the status of official languages by the Government of India. According to the 2001 census, 29 Indian languages have more than a million speakers. Most of these languages, except for Hindi, are low resource. Many of these, do not have a written script and hence, speech technology solutions would greatly benefit such communities. To be able to truly support speech and language systems that can be used by everyone in the country, we need to come up with techniques to build systems in these resource constrained settings, while also exploiting the unique properties and similarities between Indian languages.<\/p>\n<p>We are releasing data in Telugu, Tamil and Gujarati, and participants in this challenge will be required to use only the released data to build ASR systems in these languages, which will make the task fair for all participants and direct the focus of the work to the low resource setting. However, we will not restrict participants from only working on one of the components of the ASR pipeline \u2013 participants will be free to innovate in any aspect of the ASR system as long as they only use the data provided. We will release a baseline system that participants can compare their systems against and use as a starting point. During testing, we will release a held-out blind test set that the systems will be evaluated on.<\/p>\n<p><strong>Contact us:\u00a0<\/strong><a>interspeech2018@microsoft.com<\/a><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>January 2, 2018<\/strong> <\/span><span style=\"color: #000000;font-family: Calibri\">\u2013 Registration for the challenge opens, training data released<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>January 15, 2018<\/strong> \u2013 Release baseline recipe <\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 6, 2018<\/strong> \u2013 Test portal opens at 10 am IST (Indian Standard Time, GMT+05:30) with test audio released.<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 9, 2018<\/strong> \u2013 Test deadline for competition at 5 pm IST (Indian Standard Time, GMT+05:30); up to 3 hypotheses files per language<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 16, 2018<\/strong> \u2013 Abstract submission deadline<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 23, 2018<\/strong> \u2013 Interspeech final paper upload deadline<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>June 17, 2018<\/strong> \u2013 Camera ready paper deadline<\/span><\/p>\n<p>Note: While submitting the paper, please make sure you select the special session &#8220;Low Resource Speech Recognition Challenge for Indian Languages&#8221;.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p>1. To participate in the Challenge, you must register and consent to the agreement at the \u201cRegister\u201d page and download the data. Participants may not share the data with any person or organization without Microsoft\u2019s prior written consent.<\/p>\n<p>2. Participants who register but do not submit a system to the Challenge are considered withdrawn from the Challenge and are required to delete and purge all data copies.<\/p>\n<p>3. To qualify for the Challenge, Participants must submit a system created on the following guidelines:<\/p>\n<ul>\n<li>Participants may only use the audio and transcriptions provided to build their systems.<\/li>\n<li>Participants may choose to use the corresponding language\u2019s data to build each system or combine the data and use it cross-lingually.<\/li>\n<li>Participants may build systems for any number of languages, even if they all use the data.<\/li>\n<li>The systems submitted are expected to beat the baseline system in terms of WER, however, innovative systems that come close to the baseline may be considered.<\/li>\n<li>Only the audio for the blind test set (5 hours) will be released. Participants are expected to run their systems on the blind test set.<\/li>\n<li>Participants must submit the following items to Microsoft for evaluation: (1) the ASR hypotheses; (2) the final ASR model; and (3) the research paper so Microsoft can reproduce the hypotheses against the blind set.<\/li>\n<\/ul>\n<p>4. Participants who register and submit systems to the Challenge may use the data in the future solely for research purposes. Participants should provide the following attribution when they publish their findings, \u201cData provided by SpeechOcean.com and Microsoft\u201d. Data may not be used for commercial purposes.<\/p>\n<p>If you have any questions, please write to <a href=\"mailto:interspeech2018@microsoft.com\">interspeech2018@microsoft.com<\/a><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p>Registration for the challenge is now closed.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p>Baselines are built using Kaldi. Please see this <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" target=\"_blank\" href=\"https:\/\/1drv.ms\/w\/s!AvNigI3ur_6FgdEU1w97P7xHvww6_g\">README file<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> for instructions on how to replicate the baselines and for links to lexicons for all languages.<\/p>\n<p>The Word Error Rates of the baseline systems for all three languages are below:<\/p>\n<table style=\"height: 115px;border-collapse: collapse;border-spacing: inherit\" width=\"827\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\"><strong>Language<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>GMM-HMM<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>DNN<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>TDNN<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Tamil<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a033.55<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a025.47<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a019.45<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Telugu<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a040.12<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a034.97<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a022.61<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Gujarati<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a023.78<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a027.79<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a019.76<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p><strong>Language: Gujarati<\/strong><\/p>\n<p>Models submitted: 40<\/p>\n<p>Number of teams: 18<\/p>\n<table style=\"height: 78px;border-collapse: collapse;border-spacing: inherit\" width=\"580\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.06%, 14.70%, 15.04%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.69%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">ISI-Billa<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">19.31%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><strong>Language: Tamil<\/strong><\/p>\n<p>Models submitted: 36<\/p>\n<p>Number of teams: 14<\/p>\n<table style=\"height: 102px;border-collapse: collapse;border-spacing: inherit\" width=\"586\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">13.92%, 14.08%, 14.27%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.07%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.32%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><strong>Language: Telugu<\/strong><\/p>\n<p>Models submitted: 33<\/p>\n<p>Number of teams: 18<\/p>\n<table style=\"height: 41px;border-collapse: collapse;border-spacing: inherit\" width=\"588\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.71%, 14.86%, 15.07%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.14%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.59%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p>Note: Final winners will be determined after verification and replication of results.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Organizing Committee: Kalika Bali Krishna Doss Mohan Rupesh Kumar Mehta Niranjan Nayak Sunayana Sitaram Radhakrishnan Srikanth Data preparation and baselines: Brij Mohan Lal Srivastava Pallavi Matani Sandeepkumar Satpal Satarupa Guha Shambo Chatterjee Swapnajeet Padhi Opens in a new tab In keeping with the Interspeech 2018 theme of \u2018Speech Research for Emerging Markets in Multilingual Societies\u2019, [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_startdate":"2018-06-17","msr_enddate":"2018-06-17","msr_location":"","msr_expirationdate":"","msr_event_recording_link":"","msr_event_link":"","msr_event_link_redirect":false,"msr_event_time":"","msr_hide_region":false,"msr_private_event":false,"msr_hide_image_in_river":0,"footnotes":""},"research-area":[13545],"msr-region":[],"msr-event-type":[],"msr-video-type":[],"msr-locale":[268875],"msr-program-audience":[],"msr-post-option":[],"msr-impact-theme":[],"class_list":["post-451875","msr-event","type-msr-event","status-publish","hentry","msr-research-area-human-language-technologies","msr-locale-en_us"],"msr_about":"<!-- wp:msr\/event-details {\"title\":\"Interspeech 2018 Special Session: Low Resource Speech Recognition Challenge for Indian Languages\",\"backgroundColor\":\"grey\",\"imageType\":\"full-bleed\"} \/-->\n\n<!-- wp:msr\/content-tabs --><!-- wp:msr\/content-tab {\"title\":\"Challenge Overview\"} --><!-- wp:freeform --><p><strong>Organizing Committee:<\/strong><\/p>\n<ul>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/kalikab\/\">Kalika Bali<\/a><\/li>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/kdoss\/\">Krishna Doss Mohan<\/a><\/li>\n<li>Rupesh Kumar Mehta<\/li>\n<li>Niranjan Nayak<\/li>\n<li><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/susitara\/\">Sunayana Sitaram<\/a><\/li>\n<li>Radhakrishnan Srikanth<\/li>\n<\/ul>\n<p><strong>Data preparation and baselines:<\/strong><\/p>\n<ul>\n<li>Brij Mohan Lal Srivastava<\/li>\n<li>Pallavi Matani<\/li>\n<li>Sandeepkumar Satpal<\/li>\n<li>Satarupa Guha<\/li>\n<li>Shambo Chatterjee<\/li>\n<li>Swapnajeet Padhi<\/li>\n<\/ul>\n<p><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<p>In keeping with the Interspeech 2018 theme of \u2018Speech Research for Emerging Markets in Multilingual Societies\u2019, we\u00a0are organizing\u00a0a special session and challenge on speech recognition for low resource languages. Most languages in the world lack the amount of text, speech and linguistic resources required to build large Deep Neural Network (DNN)-based models. However, there have been many advances in DNN architectures, cross-lingual and multilingual speech processing techniques, and approaches incorporating linguistic knowledge into machine-learning based models, that can help in building systems for low resource languages. In this challenge, we\u00a0will focus on building Automatic Speech Recognition (ASR) systems for Indian languages with constraints on the data available for Acoustic Modeling and Language Modeling.<\/p>\n<p>India has around 1500 languages, of which 22 languages have been given the status of official languages by the Government of India. According to the 2001 census, 29 Indian languages have more than a million speakers. Most of these languages, except for Hindi, are low resource. Many of these, do not have a written script and hence, speech technology solutions would greatly benefit such communities. To be able to truly support speech and language systems that can be used by everyone in the country, we need to come up with techniques to build systems in these resource constrained settings, while also exploiting the unique properties and similarities between Indian languages.<\/p>\n<p>We are releasing data in Telugu, Tamil and Gujarati, and participants in this challenge will be required to use only the released data to build ASR systems in these languages, which will make the task fair for all participants and direct the focus of the work to the low resource setting. However, we will not restrict participants from only working on one of the components of the ASR pipeline \u2013 participants will be free to innovate in any aspect of the ASR system as long as they only use the data provided. We will release a baseline system that participants can compare their systems against and use as a starting point. During testing, we will release a held-out blind test set that the systems will be evaluated on.<\/p>\n<p><strong>Contact us:\u00a0<\/strong><a>interspeech2018@microsoft.com<\/a><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- wp:msr\/content-tab {\"title\":\"Challenge Timeline\"} --><!-- wp:freeform --><p><span style=\"color: #000000;font-family: Calibri\"><strong>January 2, 2018<\/strong> <\/span><span style=\"color: #000000;font-family: Calibri\">\u2013 Registration for the challenge opens, training data released<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>January 15, 2018<\/strong> \u2013 Release baseline recipe <\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 6, 2018<\/strong> \u2013 Test portal opens at 10 am IST (Indian Standard Time, GMT+05:30) with test audio released.<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 9, 2018<\/strong> \u2013 Test deadline for competition at 5 pm IST (Indian Standard Time, GMT+05:30); up to 3 hypotheses files per language<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 16, 2018<\/strong> \u2013 Abstract submission deadline<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>March 23, 2018<\/strong> \u2013 Interspeech final paper upload deadline<\/span><\/p>\n<p><span style=\"color: #000000;font-family: Calibri\"><strong>June 17, 2018<\/strong> \u2013 Camera ready paper deadline<\/span><\/p>\n<p>Note: While submitting the paper, please make sure you select the special session &#8220;Low Resource Speech Recognition Challenge for Indian Languages&#8221;.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- wp:msr\/content-tab {\"title\":\"Challenge Rules\"} --><!-- wp:freeform --><p>1. To participate in the Challenge, you must register and consent to the agreement at the \u201cRegister\u201d page and download the data. Participants may not share the data with any person or organization without Microsoft\u2019s prior written consent.<\/p>\n<p>2. Participants who register but do not submit a system to the Challenge are considered withdrawn from the Challenge and are required to delete and purge all data copies.<\/p>\n<p>3. To qualify for the Challenge, Participants must submit a system created on the following guidelines:<\/p>\n<ul>\n<li>Participants may only use the audio and transcriptions provided to build their systems.<\/li>\n<li>Participants may choose to use the corresponding language\u2019s data to build each system or combine the data and use it cross-lingually.<\/li>\n<li>Participants may build systems for any number of languages, even if they all use the data.<\/li>\n<li>The systems submitted are expected to beat the baseline system in terms of WER, however, innovative systems that come close to the baseline may be considered.<\/li>\n<li>Only the audio for the blind test set (5 hours) will be released. Participants are expected to run their systems on the blind test set.<\/li>\n<li>Participants must submit the following items to Microsoft for evaluation: (1) the ASR hypotheses; (2) the final ASR model; and (3) the research paper so Microsoft can reproduce the hypotheses against the blind set.<\/li>\n<\/ul>\n<p>4. Participants who register and submit systems to the Challenge may use the data in the future solely for research purposes. Participants should provide the following attribution when they publish their findings, \u201cData provided by SpeechOcean.com and Microsoft\u201d. Data may not be used for commercial purposes.<\/p>\n<p>If you have any questions, please write to <a href=\"mailto:interspeech2018@microsoft.com\">interspeech2018@microsoft.com<\/a><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- wp:msr\/content-tab {\"title\":\"Registration\"} --><!-- wp:freeform --><p>Registration for the challenge is now closed.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- wp:msr\/content-tab {\"title\":\"Baselines\"} --><!-- wp:freeform --><p>Baselines are built using Kaldi. Please see this <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" target=\"_blank\" href=\"https:\/\/1drv.ms\/w\/s!AvNigI3ur_6FgdEU1w97P7xHvww6_g\">README file<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> for instructions on how to replicate the baselines and for links to lexicons for all languages.<\/p>\n<p>The Word Error Rates of the baseline systems for all three languages are below:<\/p>\n<table style=\"height: 115px;border-collapse: collapse;border-spacing: inherit\" width=\"827\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\"><strong>Language<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>GMM-HMM<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>DNN<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\"><strong>TDNN<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Tamil<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a033.55<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a025.47<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a019.45<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Telugu<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a040.12<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a034.97<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a022.61<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\">Gujarati<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a023.78<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a027.79<\/td>\n<td style=\"padding: inherit;border: inherit\">\u00a019.76<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- wp:msr\/content-tab {\"title\":\"Leaderboard\"} --><!-- wp:freeform --><p><strong>Language: Gujarati<\/strong><\/p>\n<p>Models submitted: 40<\/p>\n<p>Number of teams: 18<\/p>\n<table style=\"height: 78px;border-collapse: collapse;border-spacing: inherit\" width=\"580\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.06%, 14.70%, 15.04%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.69%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">ISI-Billa<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">19.31%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><strong>Language: Tamil<\/strong><\/p>\n<p>Models submitted: 36<\/p>\n<p>Number of teams: 14<\/p>\n<table style=\"height: 102px;border-collapse: collapse;border-spacing: inherit\" width=\"586\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">13.92%, 14.08%, 14.27%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.07%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.32%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p><strong>Language: Telugu<\/strong><\/p>\n<p>Models submitted: 33<\/p>\n<p>Number of teams: 18<\/p>\n<table style=\"height: 41px;border-collapse: collapse;border-spacing: inherit\" width=\"588\">\n<tbody>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.71%, 14.86%, 15.07%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.14%<\/td>\n<\/tr>\n<tr>\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.59%<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>&nbsp;<\/p>\n<p>Note: Final winners will be determined after verification and replication of results.<span id=\"label-external-link\" class=\"sr-only\" aria-hidden=\"true\">Opens in a new tab<\/span><\/p>\n<!-- \/wp:freeform --><!-- \/wp:msr\/content-tab --><!-- \/wp:msr\/content-tabs -->","tab-content":[{"id":0,"name":"Challenge Overview","content":"In keeping with the Interspeech 2018 theme of \u2018Speech Research for Emerging Markets in Multilingual Societies\u2019, we\u00a0are organizing\u00a0a special session and challenge on speech recognition for low resource languages. Most languages in the world lack the amount of text, speech and linguistic resources required to build large Deep Neural Network (DNN)-based models. However, there have been many advances in DNN architectures, cross-lingual and multilingual speech processing techniques, and approaches incorporating linguistic knowledge into machine-learning based models, that can help in building systems for low resource languages. In this challenge, we\u00a0will focus on building Automatic Speech Recognition (ASR) systems for Indian languages with constraints on the data available for Acoustic Modeling and Language Modeling.\r\n\r\nIndia has around 1500 languages, of which 22 languages have been given the status of official languages by the Government of India. According to the 2001 census, 29 Indian languages have more than a million speakers. Most of these languages, except for Hindi, are low resource. Many of these, do not have a written script and hence, speech technology solutions would greatly benefit such communities. To be able to truly support speech and language systems that can be used by everyone in the country, we need to come up with techniques to build systems in these resource constrained settings, while also exploiting the unique properties and similarities between Indian languages.\r\n\r\nWe are releasing data in Telugu, Tamil and Gujarati, and participants in this challenge will be required to use only the released data to build ASR systems in these languages, which will make the task fair for all participants and direct the focus of the work to the low resource setting. However, we will not restrict participants from only working on one of the components of the ASR pipeline \u2013 participants will be free to innovate in any aspect of the ASR system as long as they only use the data provided. We will release a baseline system that participants can compare their systems against and use as a starting point. During testing, we will release a held-out blind test set that the systems will be evaluated on.\r\n\r\n<strong>Contact us:\u00a0<\/strong><a>interspeech2018@microsoft.com<\/a>"},{"id":1,"name":"Challenge Timeline","content":"<span style=\"color: #000000;font-family: Calibri\"><strong>January 2, 2018<\/strong> <\/span><span style=\"color: #000000;font-family: Calibri\">\u2013 Registration for the challenge opens, training data released<\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>January 15, 2018<\/strong> \u2013 Release baseline recipe <\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>March 6, 2018<\/strong> \u2013 Test portal opens at 10 am IST (Indian Standard Time, GMT+05:30) with test audio released.<\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>March 9, 2018<\/strong> \u2013 Test deadline for competition at 5 pm IST (Indian Standard Time, GMT+05:30); up to 3 hypotheses files per language<\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>March 16, 2018<\/strong> \u2013 Abstract submission deadline<\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>March 23, 2018<\/strong> \u2013 Interspeech final paper upload deadline<\/span>\r\n\r\n<span style=\"color: #000000;font-family: Calibri\"><strong>June 17, 2018<\/strong> \u2013 Camera ready paper deadline<\/span>\r\n\r\nNote: While submitting the paper, please make sure you select the special session \"Low Resource Speech Recognition Challenge for Indian Languages\"."},{"id":2,"name":"Challenge Rules","content":"1. To participate in the Challenge, you must register and consent to the agreement at the \u201cRegister\u201d page and download the data. Participants may not share the data with any person or organization without Microsoft\u2019s prior written consent.\r\n\r\n2. Participants who register but do not submit a system to the Challenge are considered withdrawn from the Challenge and are required to delete and purge all data copies.\r\n\r\n3. To qualify for the Challenge, Participants must submit a system created on the following guidelines:\r\n<ul>\r\n \t<li>Participants may only use the audio and transcriptions provided to build their systems.<\/li>\r\n \t<li>Participants may choose to use the corresponding language\u2019s data to build each system or combine the data and use it cross-lingually.<\/li>\r\n \t<li>Participants may build systems for any number of languages, even if they all use the data.<\/li>\r\n \t<li>The systems submitted are expected to beat the baseline system in terms of WER, however, innovative systems that come close to the baseline may be considered.<\/li>\r\n \t<li>Only the audio for the blind test set (5 hours) will be released. Participants are expected to run their systems on the blind test set.<\/li>\r\n \t<li>Participants must submit the following items to Microsoft for evaluation: (1) the ASR hypotheses; (2) the final ASR model; and (3) the research paper so Microsoft can reproduce the hypotheses against the blind set.<\/li>\r\n<\/ul>\r\n4. Participants who register and submit systems to the Challenge may use the data in the future solely for research purposes. Participants should provide the following attribution when they publish their findings, \u201cData provided by SpeechOcean.com and Microsoft\u201d. Data may not be used for commercial purposes.\r\n\r\nIf you have any questions, please write to <a href=\"mailto:interspeech2018@microsoft.com\">interspeech2018@microsoft.com<\/a>"},{"id":3,"name":"Registration","content":"Registration for the challenge is now closed."},{"id":4,"name":"Baselines","content":"Baselines are built using Kaldi. Please see this <a href=\"https:\/\/1drv.ms\/w\/s!AvNigI3ur_6FgdEU1w97P7xHvww6_g\">README file<\/a> for instructions on how to replicate the baselines and for links to lexicons for all languages.\r\n\r\nThe Word Error Rates of the baseline systems for all three languages are below:\r\n<table style=\"height: 115px;border-collapse: collapse;border-spacing: inherit\" width=\"827\">\r\n<tbody>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\"><strong>Language<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\"><strong>GMM-HMM<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\"><strong>DNN<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\"><strong>TDNN<\/strong><\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\">Tamil<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a033.55<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a025.47<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a019.45<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\">Telugu<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a040.12<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a034.97<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a022.61<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\">Gujarati<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a023.78<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a027.79<\/td>\r\n<td style=\"padding: inherit;border: inherit\">\u00a019.76<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>"},{"id":5,"name":"Leaderboard","content":"<strong>Language: Gujarati<\/strong>\r\n\r\nModels submitted: 40\r\n\r\nNumber of teams: 18\r\n<table style=\"height: 78px;border-collapse: collapse;border-spacing: inherit\" width=\"580\">\r\n<tbody>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.06%, 14.70%, 15.04%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.69%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">ISI-Billa<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">19.31%<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n&nbsp;\r\n\r\n<strong>Language: Tamil<\/strong>\r\n\r\nModels submitted: 36\r\n\r\nNumber of teams: 14\r\n<table style=\"height: 102px;border-collapse: collapse;border-spacing: inherit\" width=\"586\">\r\n<tbody>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">13.92%, 14.08%, 14.27%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.07%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">16.32%<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n&nbsp;\r\n\r\n<strong>Language: Telugu<\/strong>\r\n\r\nModels submitted: 33\r\n\r\nNumber of teams: 18\r\n<table style=\"height: 41px;border-collapse: collapse;border-spacing: inherit\" width=\"588\">\r\n<tbody>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\"><strong>Team Name<\/strong><\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\"><strong>Word Error Rate<\/strong><\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Jilebi<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">14.71%, 14.86%, 15.07%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">Cogknit<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.14%<\/td>\r\n<\/tr>\r\n<tr>\r\n<td style=\"padding: inherit;border: inherit\" width=\"125\">CSALT-LEAP<\/td>\r\n<td style=\"padding: inherit;border: inherit\" width=\"180\">17.59%<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n&nbsp;\r\n\r\nNote: Final winners will be determined after verification and replication of results."}],"msr_startdate":"2018-06-17","msr_enddate":"2018-06-17","msr_event_time":"","msr_location":"","msr_event_link":"","msr_event_recording_link":"","msr_startdate_formatted":"June 17, 2018","msr_register_text":"Watch now","msr_cta_link":"","msr_cta_text":"","msr_cta_bi_name":"","featured_image_thumbnail":null,"event_excerpt":"In keeping with the Interspeech 2018 theme of \u2018Speech Research for Emerging Markets in Multilingual Societies\u2019, we\u00a0are organizing\u00a0a special session and challenge on speech recognition for low resource languages. Most languages in the world lack the amount of text, speech and linguistic resources required to build large Deep Neural Network (DNN)-based models. However, there have been many advances in DNN architectures, cross-lingual and multilingual speech processing techniques, and approaches incorporating linguistic knowledge into machine-learning based&hellip;","msr_research_lab":[199562],"related-researchers":[],"msr_impact_theme":[],"related-academic-programs":[],"related-groups":[],"related-projects":[],"related-opportunities":[],"related-publications":[],"related-videos":[],"related-posts":[],"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event\/451875","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-event"}],"version-history":[{"count":11,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event\/451875\/revisions"}],"predecessor-version":[{"id":1147129,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event\/451875\/revisions\/1147129"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=451875"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=451875"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=451875"},{"taxonomy":"msr-event-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event-type?post=451875"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=451875"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=451875"},{"taxonomy":"msr-program-audience","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-program-audience?post=451875"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=451875"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=451875"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}