Intelligibility of speech spoken in noise and reverberation

When we speak in an environment with noise, we often modify our speech production, and speech spoken in noise is generally more intelligible than speech produced in a quiet environment, which is known as the Lombard effect. Our goal is to provide intelligible speech announcements in noisy and/or reverberant public spaces, such as train stations. Thus, the present paper examines whether speech spoken in reverberation as well as in noise improves speech intelligibility than speech spoken in quiet. We recorded words in a carrier sentence produced by two native speakers of Japanese in quiet, white noise, and reverberation (reverberation time of 3.6 s). We added white noise to the recorded speech at a signal-to-noise ratio of -2 dB or convolved it using one of two impulse responses (reverberation times of 3.6 and 2.6 s). Then we conducted listening tests with the speech sounds for 32 young native speakers of Japanese. The results show that the speech spoken not only in noise but also in reverberation is more intelligible than speech spoken in quiet. The intelligible reverberation-induced speech is observed regardless of whether speakers and listeners were in the same or different reverberant situations. The results further show that modification of the pre-target phrase contributes more to the improvements of speech intelligibility in reverberation than modification of the target word.