The Czech team of Jan Černocký BUT Speech@FIT collaborates on projects which help to identify and transcribe speech and sound, Facebook, as well as the American Defense Advanced Research Projects Agency (DARPA).
Scientists from #brnoregion collaborate with the American Defense Advanced Research Projects Agency (DARPA) to improve technologies designed to identify speeches in low quality records, e.g. terrorists using amateur radios to arrange for a meeting. The sound and signal of such communication is much worse than of a standard phone.
The Czech team of Jan Černocký alongside other teams from all over the world has been involved in the project RATS, in speaker identification, language identification and key word spotting technology.
“To make the long story short, we get a record not prepared for an IT processing. We try to get as much information as possible from the record. In case of two speakers involved, we identify who is speaking and when. We identify moments of silence and speech activities. This may seem easy but we sometimes get records from radios with noise, buzz and rustle which makes the job a real challenge,” says Černocký.
How It All Began
There is a great history of acoustic and sound research in #brnoregion. Jan Černocký stood at the beginnings of the group BUT Speech@FIT 20 years ago. As a PhD student, Černocký started working in then just founded signal processing laboratory, focusing on speech processing. Having got the PhD decree, he started collaboration with French colleagues who invited the Czech team into the first European project SpeechDat-East which was supposed to produce five language databases for voice driven teleservices.
Later on, thanks to Hynek Heřmanský, the team of Mr. Černocký went to the USA where they participated in a few significant projects.
Mr. Hynek Heřmanský is a world-renowned authority in speech processing. He works at the Center for Language and Speech Processing of John Hopkins Whiting School of Engineering, however his career started in #brnoregion.
In 2005 the group made their global breakthrough. They taxed their experience in the international evaluation of speech recognition systems and they came up with best results. According to Černocký, that was when everybody suddenly became interested in what and where Brno is, and what BUT stands for.
A year later they won even more respect for speaker recognition, and began collaboration in projects financed by the US government (which continues up to the present). That is one of the reasons they have not ceased in their efforts: the teams are continuously assessed and those with poorest results leave the competition.
Masters of Sound Processing in World of Commerce
The academic research became practical when the group started collaboration with the Czech Ministry of Defence. The University could not supply production technologies so the team of Mr. Černocký started the company Phonexia specializing in speech processing technologies up to now.
Phonexia has developed voice recognition systems in over eight languages. Their speech technologies are employed in great banks, military and police forces of West Europe and greatest European call centres. The system in a bank identifies the voice of a caller, therefore the banker does not have to ask for identification data.
The call centres use the key word identification monitoring to identify which products are most often discussed, which are the most frequent questions and if the operators do good jobs.
The group BUT Speech@FIT created an app SpokenData.com whose skills may not be all that fantastic, however they are extremely useful. You can upload a sound record and get its transcript in a moment. Journalists and many other will enjoy working with this application as it saves time spent on transcription of interviews etc.
“As for dictation into a computer, there is a little room for improvement. However, speech recognition in distorted records or in foreign languages, there’s enough to keep me busy until I retire,” Černocký admits.
AI for Facebook
Even Facebook took notice of Brno University of Technology (BUT). They chose the university among 15 elite universities granted efficient servers to speed up their research of artificial intelligence. The team of BUT shall use them to accelerate processing data acquired from speech. We all are really curious what the expert from #brnoregion come up with next!