Mathematics for the ears: Bochum engineers improve signal processing in hearing aids

Bochum researchers develop new algorithms for signal processing in hearing aids, the hearing-impaired persons to communicate by background noise or reverberation in the future easier. Furthermore, they report in RUBIN, the scientific journal of the Ruhr University. The team of Prof. Dr. Rainer Martin of the RUB-Chair of Information Technology and Communication Acoustics separates voice signals and background noise on the basis of the so-called high-resolution Kurzzeitspektralanalyse. About special calculations of useful sound is then amplified and dampened the noise. "The quality and clarity of language can be significantly improved by these measures," said Prof. Martin. "Our research is now more ways to facilitate future hearing with a hearing aid in difficult acoustic situations."

More language, less noise

Many hearing aids dampen noise, such as road noise by filtering out noise from a particular direction, since most noise coming from behind or from the side. There remain, however, unwanted noise from other directions will have to be damped by additional signal processing. To improve the acoustic signal on, then the algorithm by Prof. Martin and Dr. Timothy Gerken (Sound and Image Processing Lab, KTH Stockholm), a step further. Researchers identify both the noise and the speech components in the acoustic signal. Through a special weighting function, the Wiener filter, it can dampen the noise and stress so the speech signal.

Language found in the signal chaos

Because during any communication between the spoken syllables and words always short pauses, there are places in the signal, where the only noise is heard. The algorithm of Professor Martin's team performs a spectral analysis, ie, the acoustic signal is divided into short time segments, each of which contained all the frequencies are determined. Then those pieces signal had been identified, including only the noise. From this information, the noise spectrum can continuously in the acoustic signal is determined and also the voice portion to be estimated. For each data point, the algorithm then computes a weighting function for speech and noise signal. He finds a compromise in order to reduce noise as much as possible without distorting the speech signal.

The algorithm can more

The weighted signal that is played over loudspeakers in the instrument is modified by the Bochum algorithm further. Artifacts that arise in the estimation of the speech signal can lead to different sounds ("musical noise") that are experienced by hearing aid users as a nuisance. Professor Martin's team has designed the algorithm so that these artifacts can be suppressed, without compromising quality of voice signal.

Source: RUHR University Boschum