Investigation of the optimum acoustical conditions for speech using auralization

Abstract

i n t r o d u c t i o nSpeech intelligibility is known to be mainly determined by the signal-to-noise level difference at a listener, and reverberation. It is directly related to signal-tonoise level difference and is inversely related to the reverberation time. However, their interaction results in a complicated situation in real rooms. Increased reverberation benefits speech intelligibility by increasing speech levels. Noise is also affected by the reverberation time, as are the speech levels. Thus, the spatial relationship between a listener and the sound sources -both speech and noiseaffects the optimal reverberation for speech intelligibility in rooms [1].In this project, an experimental approach to identify the optimal reverberation time in an idealized room, and validate theoretical prediction, considering babble noise sources inside the room, is presented using auralization. Realistic optimal reverberation times are found using speech-intelligibility tests with normal-hearing and hard-ofhearing subjects. The best metric predicting speech intelligibility is presented for both normal-hearing and hardof-hearing groups. METHOD SubjectsSubject groups for the study were normal-hearing and hard-of-hearing people with a mild to moderate sensori neural hearing loss. The hearing-loss criterion for the hardof-hearing subjects was less than 20 dB at 250 Hz and 500 Hz and greater than 30 dB at 1 kHz to 8 kHz. Twenty-four normal-hearing subjects (mean age=27) and ten hard-ofhearing subjects (mean age=58) completed the tests. Test materials and simulated sound fieldsThe Modified Rhyme Test (MRT) was used as the speech-intelligibility test. The MRT lists were processed through the CATT-Acoustics room-acoustical prediction and auralization program with 4-talker babble noise The volume of the virtual room used for the simulation was 385 m3 (11 m x 7 m x 5 m). Fig. 1 shows the elevation of the virtual room, and the relative positions of the listener, the speaker, and the noise sources. In this study, the object was to model an idealized room with exponential sound decay. The effects of the different distributions of various surface materials on the walls, floor and ceiling were excluded by using the same absorption coefficients and diffusion coefficients for all octave bands and for all of the surfaces.A total of 16 different sound fields were created, consisting of the combinations of 2 different speech-and noise-source output levels (SNS = 0 and +5 dB), 4 different reverberation times (RT = 0, 0.2, 0.4, and 0.8 s), and 2 different positions of the noise source (see Fig. 1). The speech-to-noise level difference at the listener (SNR) varied from -6 dB to 8.5 dB.The completed auralization test materials were transferred to a compact disc for presentation using a CD player. Hearing screening tests were done prior to the main speech-intelligibility test to identify the hearing categories of subjects. The tests were processed through a Sony MDR V600 headphone in a soundproof room. RESULTSFig . 2 shows the mean speech-intellig...