Speech technologies in #brnoregion

At the turn of the millennium, no one from the world of speech technology had ever heard of Brno. That all began to change five years later. Now almost the entire world knows that #brnoregion and speech technologies go hand in hand.

In the 1990s, Jan Černocký, a postgrad student at Brno University of Technology, returned from his studies in France with a new passion. At the time, speech technologies were quite a new and interesting discipline, so it wasn’t difficult for him to get his colleagues and students excited about them. In 1997, four enthusiasts came together to create what is today’s BUT Speech@FIT group, and since then its membership has gradually grown to include dozens more, putting the Czech Republic on the map in the speech technologies discipline.

The group’s journey to join the elite began mainly in 20052006 when it achieved excellent results at the international evaluation of speech recognition systems organized by the American NIST. “First, we received a recording to determine which language it was in. At the time, we were just rookies and people couldn’t even pronounce Brno let alone point to it on a map, so they were rather suspicious. In 2006, though, we were successful once again, and people from all over the world started to notice us,” remembers Jan Černocký, today’s head of the BUT Speech@FIT research group, as well as professor and head of the Department of Computer Graphics and Multimedia at BUT.

Apart from academic researchers, numerous companies based in #brnoregion contribute to the great reputation of speech experts from South Moravia. And two of them – Phonexia and ReplayWell – got their start as spin-offs at BUT, still benefit from that cooperation. “We’ve been cooperating on several projects, sharing our knowledge, as we may be better at some things while the university excels at other things,” says Michal Hrabí, CEO at Phonexia, about their cooperation. 

Crime investigation and psychology

Speech technologies are particularly useful in the security and defence industries. After all, one of the strongest supporters of its development early on was the DARPA agency from the American Defense Ministry. 

That’s one of the reasons why the crews at both BUT and Phonexia deal in defence or crime investigation technologies today. Using technologies from Phonexia, detectives no longer have to spend long hours going through recordings to find incriminating evidence. “Our technology highlights the recording the detectives should listen to,” says Hrabí.
 

Among their major clients are contact centres of banks, insurance companies, and energy providers where speech technologies help supervisors evaluate call quality. The supervisors receive an aggregated report on what happened during each monitored call, such as information about its duration as well as content or emotions. “Typically, a supervisor in a contact centre can only listen to around 3-4% of all the calls; using our technologies, though, they can monitor 100% of calls by having the recordings prioritized, quickly going through the calls’ transcriptions, and listening to isolated recording or excerpts,” explains Hrabí. 

 

The technologies from Phonexia help in a lot more areas, too. Speaker authentication based on voice patterns helps contact centres verify the customers or find out whether the operator really is their employee, and they also help developing African countries provide services via phone that require caller identification.

Pavel Matějka from Speech@BUT is working on a project to help therapists evaluate therapeutic processes. The goal of this project, a joint venture between researchers from BUT and their colleagues at Masaryk University, is to create a tool to enable automated analysis of audio recordings of therapy sessions. Recognition and analysis of speech make it possible to show how much time the client or therapist spent talking, who interrupted whom, and whether they both use a similar vocabulary (helpful in confirming mutual understanding). Another project aiming to use speech technologies in emergencies, such as when a tornado hit South Moravia not long ago, is worth mentioning. The sheer volume of call activity during crises like this can cause traditional hotlines to collapse, but hotlines that include speech technology could implement a voice-based chatbot to dispense information effectively and reduce strain on human operators.

Interspeech? A must-go conference! 

Word of #brnoregion’s prowess in speech technology has spread all over the world, as proved by the international scientific conference Interspeech, which took place in Brno in 2021 and received great acclaim. Interspeech is the biggest scientific conference focusing on speech technologies and takes place in a different country every year.
 

Hosting the conference in Brno is a huge sign of appreciation for the local speech group of Jan Černocký. “It’s THE event, just like the Olympics. A must-go thing,” adds Michal Hrabí.
 

Hopefully, the conference will motivate more people to study speech technologies or technical disciplines in general, which is something Černocký would very much appreciate. “Of course, speech technology is a demanding, yet fun discipline, with great potential, and it fully trains students for work with machine learning as well, which is in high demand at numerous companies today. Technical disciplines open the door to the entire world, no matter if you work for a corporation or a small company or you founded your own start-up.”