Building a groundbreaking speech recognition system
Professor Aditi Lahiri, Emerita and Research Professor of Linguistics and Director of the Language and Brain Laboratory, has spent many years carrying out cutting-edge research into how the human brain processes language. Now, her longstanding commitment and curiosity has led to the creation of an innovative new technology with exciting possibilities for the future.
After being awarded two European Research Council grants, Professor Lahiri and her team at the Language and Brain Lab (LAB) began to develop an automated speech recognition system that was based on her model of how the brain processes sounds.
From the beginning, Professor Lahiri’s research into phonological features underpinned their work. She hypothesised that the brain recognises general patterns, and so can accept variation across sounds. As a result, a responsive speech recognition system should be able to extract features from speech sounds and match them with a defined lexicon.
“The idea was, can we make machines match the human brain? This is not what automatic speech recognition systems usually do — they don’t usually think about the human brain,” she explains.
The result is FlexSR (Flexible Speech Recognition System). The system, built from scratch by Professor Lahiri and her team over the course of six months, does not require the extensive training that speech recognition systems usually do, and can be easily adapted to different speakers, dialects, and languages. Her third ERC/EPSRC grant is using the system to investigate non-native speakers’ errors in L2 speech.
The team at the Language and Brain Lab have now built a prototype language learning app. Using the FlexSR technology, the app analyses words and sentences spoken into the app by the user, and then provides detailed feedback. Used in this way, language learners can receive personalised responses to improve their pronunciation.
The possibilities for language learning tools are exciting, but the FlexSR technology could be used in countless other ways. Beyond the sphere of education, FlexSR could be used in healthcare settings to help patients who are re-learning how to speak, or in any number of other environments that rely on voice-activated technology. Professor Lahiri hopes that the technology will make a positive impact.
While FlexSR is at the very cutting-edge of research innovation, Professor Lahiri says that she didn’t set out to innovate. “It’s not something I thought of one day, and said, I want to innovate,” she says. Instead, the project came from a longstanding commitment to academic research, and collaboration with colleagues and researchers to problem-solve and test long-running hypotheses. “In the end, this felt like an obvious thing that you could do,” Professor Lahiri says. The exceptional potential of FlexSR demonstrates how academic problem-solving can lead to powerful and exciting new technologies.
