Filter all offers by subject
A strong technological trend, particularly in robotics, in household appliance controls (smart home), and in the field of personal digital assistants, is voice control. This trend requires manufacturers to constantly improve the recognition rates of automatic voice recognition systems. It is essential to eliminate additive interfering signals here. Conventional solutions rely on training the recognition system with regard to the interfering sound, which is inflexible. The use of sound filters based on the weighting of spectral amplitudes is also known, although this distorts the target signal. The use of a microphone array and a beam shaper ultimately requires several microphones, the greater effort is often disproportionate and not in keeping with the form factors of the equipment.
Adaptive filters offer greater flexibility and better precision. Depending on the estimated signal-to-noise ratio (SNR) of the incoming signal, the step size of the adaptation is varied. The filters currently available, however, distort the target signal.
It would be desirable to have a solution that exploits the properties of the interfering signals in an optimum way.
Keywords: Spracherkennung, Rauschunterdrückung, Roboter, Konsumelektronik, industrielle Steuerung, Speech recognition, noise reduction, robotics, consumer electronics, industrial control
Researchers at Ruhr-Universität Bochum have now achieved this aim. According to the well-known adaptive line enhancer (ALE) principle, this filter is operated with just one single disrupted incoming signal. As not only the amplitude but also the phase of interference is taken into account, the ALE principle enables a distortion-free reduction in the interfering components. ALE is implemented as a normalized least mean square (NLMS) algorithm in individual frequency sub-bands. The adaptive stepsize control can thus be adapted to the existing signal components in any sub-band.
The new development consists of evaluating mutual information (MI) between the incoming signal and a delayed version of the incoming signal, both in the amplitude and in the phase spectrum.
- Optimal use of the spectral properties of the interfering signal
- Distortion-free reduction of interfering components
A patent has been filed for the Smart Adaptive Enhancer at the European Patent Organisation. We will be happy to inform you about the status of the procedure. The functionality of the new development has been shown in a software prototype. The first promising results for applications in robotics also already exist. On behalf of Ruhr-Universität Bochum, PROvendis offers licenses for the patent application
J. Taghia, R. Martin “A Frequency-Domain Adaptive Line Enhancer With Step-Size Control Based on Mutual Information for Harmonic Noise Reduction”, IEEE/ACM Transactions on Audio, Speech, and Language Processing (Volume: 24, Issue: 6, June 2016).