Revisiting perceptual compensation for effects of reverberation in speech identification

Abstract

Listeners were given the task to identify the stop-consonant ͓t͔ in the test-word "stir" when the word was embedded in a carrier sentence. Reverberation was added to the test-word, but not to the carrier, and the ability to identify the ͓t͔ decreased because the amplitude modulations associated with the ͓t͔ were smeared. When a similar amount of reverberation was also added to the carrier sentence, the listeners' ability to identify the stop-consonant was restored. This phenomenon has in previous research been considered as evidence for an extrinsic compensation mechanism for reverberation in the human auditory system ͓Watkins ͑2005͒. J. Acoust. Soc. Am. 118, 249-262͔. In the present study, the reverberant test-word was embedded in additional non-reverberant carriers, such as white noise, speech-shaped noise and amplitude modulated noise. In addition, a reference condition was included where the test-word was presented in isolation, i.e., without any carrier stimulus. In all of these conditions, the ability to identify the stop-consonant ͓t͔ was enhanced relative to the condition using the non-reverberant speech carrier. The results suggest that the non-reverberant speech carrier produces an interference effect that impedes the identification of the stop-consonant. These findings raise doubts about the existence of the compensation mechanism.

Perceptual compensation for the effects of reverberation on consonant identification: Evidence from studies with monaural stimuli

“…The silent-context conditions in our experiment were intended to further investigate findings of Nielsen and Dau (2010) and Watkins and Raimond (2013) where test-words from the "sir-stir" continuum were preceded by silence. The "modulation masking" theory suggested by Nielsen and Dau proposed that the dip cueing the [t] in a reverberant "sir-stir" continuum test-word could be masked by a preceding context, provided that that context contained a sufficient degree of modulation.…”

“…The context phrase preceding the test-word was subjected to three different treatments: Near-distance reverberation and far-distance reverberation (replicating conditions in experiment 1), and a silencing treatment which removed the preceding context cues and gave conditions similar to those presented to listeners in Watkins and Raimond (2013) and Nielsen and Dau (2010). The test-word itself was first reverberated at the near or far room distance as before, and was subsequently gated in some conditions following the method of Watkins and Raimond.…”

Statistics of natural reverberation enable perceptual separation of sound and space

“…Although reverberation that is unusually pronounced can degrade speech intelligibility (34), humans on the whole are remarkably robust to the profound distortion reverberation imposes (4,(10)(11)(12)(13)(14)(15). Comparable robustness remains beyond the capability of automatic speech recognition, the performance of which deteriorates under even moderate reverberation (42).…”

“…On the other hand, reverberation provides information about the environment, because reflection characteristics depend on the geometry of the space around us and the position of a sound source within it. Biological organisms are well adapted to reverberation, using it to infer room size and source distance (6)(7)(8)(9) while retaining a robust ability to identify sound sources despite the environmental distortion (10)(11)(12)(13)(14)(15). It remains unclear how the auditory system achieves these capabilities.…”

“…Because the listener lacks direct access to either the source or the filter, their estimation is ill-posed. Although the physics of reverberation are well established (2,4,5,17), as is the fact that human listeners are relatively robust to distortion from reverberation (4,(10)(11)(12)(13)(14)(15), the underlying perceptual mechanisms have been little studied and remain poorly understood. One might suppose that robustness simply results from learning how the structure of familiar sounds such as those of speech is altered under reverberation.…”