Auditory information is an important sensation for human in daily life.
Voice information and speech recognition system are key factors to realize hands-free system and an intuitive interface for home-use robots.
Although current speech recognition system can recognize humans' speech in a silent environment, it still has a difficulty in recognizing humans' speech in noisy environments. The limitation makes the applications of speech recognition system to a home-use robot difficult.
Noise reduction is a key factor to make noise-robust speech recognition system.
Mitsuharu Matsumoto laboratory progresses our research on acoustical signal processing and musical signal processing such as:
- Noise reduction combining microphones and piezoelectric devices
- Single-channel noise reduction using time-frequency ε-filter
- Aggregated microphones to minimize microphone array
Our laboratory aims to understand the mechanism of auditory sensation and apply our system to speech recognition system, robot audition and biometric authentication.
An acoustical array combining microphones and piezoelectric devices
This research aims to reduce internal noise of the robot inside by using an acoustical array combining micropjones and piezoelectric devices. As piezoelectric device can detect object vibration directly, it is expected that effective noise reduction can be realized.
Single-channel noise reduction using time-frequency ε-filter
We are studying time-frequency ε-filter to handle not only small amplitude noise but also large amplitude noise. Time-frequency ε-filter is an advanced ε-filter in time-frequency domain. The developed filter can reduce large amplitude noise, while preserving speech information.
A study on miniaturizing microphone array using aggregated microphones
We are studying a small size microphone array labeled aggregated microphones. In aggregated microphones, we set microphones located not in the different places but in the same place. It does not use omni-directional microphones but directional microphones unlike a typical microphone array. It is expected that our method can be applied to a small size robot due to its feature.