Conventional Automatic Speech Recognition systems solely rely on acoustic information, making them susceptible to problems like environmental noise, privacy, information disclosure and also excluding users with speech impairments. An Ultrasonic Doppler Sensing (UDS) based interface may be used to tackle these issues since it does not rely on audio signal information. This paper describes the first speech recognition experiments based on UDS for European Portuguese (EP). The work here presented analyzes the UDS signal and explores the recognition of EP digits and minimal pairs of words that only differ on nasality of one of the phones. The results of our experiments show a best word error rate of 27.8% using data collected with the device at different distances from the speaker in an isolated word recognition problem. Communications in Computer and Information Science book series (CCIS, volume 328)

--

Ultrasonic doppler sensing,Silent speech,European Portuguese,Nasality

• Matemáticas - Ciências Naturais
• Ciências da Computação e da Informação - Ciências Naturais