Používáte starou verzi internetového prohlížeče. Doporučujeme aktualizaci na nejnovější verzi. Některé funkce portálu nemusí být podporovány.
Pravděpodobně máte vypnutý JavaScript. Některé funkce portálu nebudou funkční.
The world's leaders in speech processing are heading to Brno. They will show whether coronavirus infection can be detected from the voice
Siri, Alexa, Google - we talk to them all today. But who breathed life into them? Speech recognition has gradually spread from IT labs to smart homes and has become a common part of everyday life. Getting information from audio recordings is becoming increasingly important. For the first time in its history, the Czech Republic will host the international conference Interspeech, which will take place in Brno from August 30 to September 3, 2021. Hundreds of world experts will arrive in the Moravian metropolis. More than 1400 will join the event online. Specialists will present news from the field, such as how to start using machine learning with a minimum of training data, or whether coronavirus infection can be recognized from the cough.
„While before the pandemic, we analysed various forms of cough to diagnose chronic lung disease, tuberculosis or asthma, for the last two years we have focused on coronavirus sound," Siriam Ganapathy said, an Indian expert, whose aim is to save testing costs and speed up diagnosis. There are two program sections on the topic related to the pandemic at Interspeech. For a long time, speakers have been obtaining information from audio recordings about the patient's health condition or mental state.
„The motto of the conference is Speech Everywhere. Whereas previously only a few groups were involved in speech data mining research, today speech recognition is included in every smartphone and many other devices. The fridge, the car and the call center voicebots are talking to us. But we are still not done with research and development. A big topic of this time is the training of machine learning models on only a small amount of data. Obtaining data from audio recordings is quite demanding and costly, so now research teams from all over the world are trying to learn how to learn only from small databases that can be easily downloaded from YouTube or the Internet in general," one of the main organizers, Honza Černocký from the Faculty of Information Technology at BUT said. Honza has been leading the BUT Speech@FIT research group for years. Illustration photo | Autor: Pixabay Among the keynote speakers will be Hermann Ney from the German University of Aachen, who will receive a medal at the conference this year for the lifelong contribution of the field from the International Speech and Communication Association ISCA. Czech science will be represented, for example, by FIT BUT graduate Tomáš Mikolov, and renowned scientists Pascale Fung from Hong Kong and Mounya Elhilali from Johns Hopkins University in the USA will also speak at the event.
„Data obtained from audio recordings is constantly being refined. Gradually, scientists from all over the world are trying to teach artificial intelligence to work with sounds even in degraded conditions, such as talking to people in the distance, with a reverberation in a church or in a restaurant full of people. We also need to address language coverage, a lot of development has been done in commercially important languages, such as English, French, German and so on, but in the future, it will be necessary to focus on less represented languages, which often do not even have a written form, only spoken,” Černocký added, who worked intensively with his colleague Lukáš Burget from FIT BUT to organize the event.
Brno started trying to organize the conference for the first time in 2015, also because a number of IT companies dedicated to the processing of audio data operate in the Moravian metropolis. For the first time, Interspeech 2021 will combine in-person and online forms, with over 350 scientists visiting Brno in-person, and another 1,400 joining the event virtually. In total, all continents will be represented, with most scientists from Europe, North America and Asia. The organizing team selected 992 scientific articles from more than two thousand submitted for presentation.
