Mitsubishi Electric has developed a breakthrough noise-suppression technology that significantly improves the quality of hands-free voice communication in noisy conditions, such as making a voice call via a car navigation system. The technology improves voice clarity by removing 96% of ambient sound, including rapidly changing noise from turn signals or wipers, which are difficult to suppress using conventional methods. While it is very difficult to develop a model that can precisely handle all noises, human speech can be clearly characterized in terms of airflow from the lungs, vocal cord vibrations and the shape of the vocal tract. In tests, Mitsubishi Electric’s new system suppressed 96% of the ambient noise, compared to only 78% for a conventional system.
Mitsubishi Electric’s powerful speech model exploits such characteristics, and incorporates a machine-learning method based on artificial neural networks that distinguish clearly between speech and non-speech sounds. This breakthrough noise-suppression technology significantly improves the quality of hands-free voice communication compared to conventional methods. Using massive amounts of data, the system learns how to process both slowly and rapidly varying noises typically present inside a car, including external noise from the road and other cars and internal noise from the air conditioner, turn signals and windshield wipers. Mitsubishi Electric aims to launch its new technology in car navigation systems in 2018.
