Measuring and modeling speech intelligibility in real and loudspeaker-based virtual sound environments

Area of Science:

• Acoustics
• Psychoacoustics
• Virtual Reality

Background:

• Loudspeaker-based virtual sound environments are crucial for studying speech perception.
• Evaluating reproduction methods for perceptual accuracy in these environments is limited.
• This study compares real rooms to simulated virtual acoustic environments.

Purpose of the Study:

• To compare speech intelligibility in a reference room versus its virtual simulations.
• To evaluate different loudspeaker reproduction methods for virtual acoustics.
• To assess the impact of reverberation, spatial configuration, and interferer type on speech intelligibility.

Main Methods:

• Compared speech intelligibility in a reference room with three virtual reproduction methods: mixed-order ambisonics, nearest loudspeaker mapping, and measured impulse responses with ambisonics.
• Varied reverberation, spatial configuration (colocated, separated interferers), and interferer type (speech, noise).
• Utilized a binaural speech intelligibility model to analyze results.

Main Results:

• Simulation-based methods (ambisonics, nearest neighbor) differed from the reference room when target and interferers were spatially separated.
• No significant differences were observed when target and interferers were colocated.
• The reproduction method using measured impulse responses and ambisonics showed the highest similarity to the reference room's speech intelligibility.

Conclusions:

• The choice of reproduction method significantly impacts the perceptual accuracy of virtual sound environments.
• Measured impulse response reproduction offers higher fidelity for speech intelligibility tasks.
• Binaural speech intelligibility models can effectively predict and explain differences in virtual acoustic environments.