Beta 1


Title Effects of compression in hearing aids on the envelope of the speech signal, Signal based measures of the side-effects of the compression and their relation to speech intelligibility
Author Walaszek, Justyna
Supervisor Dau, Torsten (Acoustic Technology, Department of Electrical Engineering, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Johannesson, René Burmand
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2008
Abstract A research project is presented, where the influence of the compression in hearing aids on the envelope of the speech signal was investigated. Signal-based measures were implemented to quantify three types of the side-effects of the compression: an across source modulation correlation (ASMC), an envelope alternation by the envelope difference index (EDI) and a change of the target-to-background ratio (TBR) at the output of the compression system. The measures were implemented for the setup, where a hearing-aid device was used. The dependence on the compression ratio, the release time and the compression strategy was studied, whereby compression was applied either before or after mixing foreground and background stimuli. The target speech was presented in a background speech or a speech-like noise. ASMC occurs when two fluctuating signals are subjected to compression and obtain a common modulation due to the compressor action. The ASMC measure was calculated for the unprocessed foreground and background as well as when the signals were compressed independently before mixing (INDEP) or compressed together after mixing (COM). As expected from data reported by Stone and Moore (2007, 2008), a negative correlation was observed only in the COM condition for the compression system with a short release time. The difference in temporal contrast of the speech envelope before and after the compression was quantified with the EDI. It was observed that the EDI increased with increasing compression ratio and decreased with increasing release time. Additionally, it was observed that the values are generally higher for the background of the one-talker speech than for the background of 2- talker modulated speech-shaped noise. The TBR was measured at the output of the compression system. It occurred that, despite of identical input TBR, the output TBR was of different value for various compression settings. The speech intelligibility test for situations, where mainly the envelope cues were preserved, was performed for various parameters of the compression and two types of fluctuating background. In the perceptual test, the intelligibility of speech was measured in the presence of background speech and speech-like noise, similarly as for the measurements of the signal-based measures. The signals were processed by an 8-channel noise vocoder, such as mainly the envelope cues were remaining. The experiment was performed with normal hearing subjects. For both types of background there was a significantly worse performance for the fast-acting compression system. For the speech-like noise background, there was also a significant interaction between the speed and the compression strategy (compression before versus after mixing foreground and background) and the pair-wise comparison of Tukey HSD test revealed that when the mixture of the foreground and background were processed with fast-acting compression system, the recognition was significantly worse. The speech intelligibility scores were compared with the pattern of signal-based measures. The observed effects could be attributed to the side-effects of the compression. These observations suggest that the signal-based measures could be relevant for predicting the intelligibility pattern for target speech presented in the presence of speech or speech-like noise that was processed with a compression hearing aid.
Pages 97
Fulltext
Original PDF JustynaWalaszek_Thesis_printN.pdf (1.19 MB)
Admin Creation date: 2009-06-09    Update date: 2009-06-09    Source: dtu    ID: 244383    Original MXD