A local company is working with researchers in CMP to develop a smart-phone app that can detect emotion in a voice signal.  EI Technologies of Saffron Walden are developing an app that can "listen" all the time to its owner's voice and extract features from the audio signal, which can then be analysed off-line to determine the speaker's emotional state at any time.

These features include things such as the pitch, intensity and duration of speech sounds and are known as paralinguistic features. The target application for this technology is people who are undergoing treatment for depression.  They are usually asked to keep a diary of their emotional state, but many people find this very difficult to maintain.  By monitoring their voice throughout the day, their physician can pinpoint exactly when they felt depressed, angry, sad etc. and discuss with them what external (or internal) factors led to this state: this information is vital in understanding the causes of their condition.

 

Prof. Stephen Cox, who is acting as scientific adviser to the company, said that paralinguistics is currently a hot area of research in computer speech processing.  "Interest in this area grew considerably in 2009 when there was an 'Emotion Challenge' workshop at an international conference in Brighton, UK, at which researchers defined tasks and data to be used in these studies.  The technology is possible because of advances in machine-learning algorithms, which enable computers to discover automatically what features in a signal are important for discriminating between different emotions."  Crucially, there are no privacy issues involved, because the features collected by the app whilst it is “listening” cannot be used to reconstruct the speech.

 

EI Technologies and Prof. Cox have been awarded a Technology Strategy Board grant of £150k to develop the technology further.