Effective speech recognition makes life easier for people and machines
“Machines will eventually learn to communicate with us verbally, as long as speech recognition technologies are designed to adjust to the noisy conditions of the real world,” Antti Hurmalainen says.
The researchers in the Department of Signal Processing have developed mathematical methods that help people and machines distinguish different sounds in noisy environments.
One of the researchers is Antti Hurmalainen, who defended his doctoral dissertation titled “Robust Speech Recognition with Spectogram Factorisation” at TUT in early October 2014.
Automatic speech recognition allows us to control machines through voice commands. Among other things, voice-controlled technology improves traffic safety. Drivers can keep their hands on the wheel and their eyes on the road while searching for information on the web.
However, users get frustrated if their voice commands are not understood.
”Hearing what others are saying in noisy environments is difficult for us people, but the same is true for machines. The development of robust speech recognition technology is challenging, because nowadays mobile devices are used everywhere,” Antti Hurmalainen says.
The systems presented in Hurmalainen’s dissertation have yielded successful results in international comparisons on which major mobile service providers keep a keen eye. In addition to the content of speech, the systems are capable of distinguishing the voices of multiple simultaneous speakers.