Speech bandwidth extension and vocal effort coding for enhanced communication in noisy industrial environments

Abstract

Communicating in noise is a difficult task for people wearing Hearing Protection Devices (HPD) in noisy industrial environments. In fact, the number one reason why workers choose not to wear HPDs in the workplace is that HPDs are a barrier to communication. Often, workers have to choose between protecting their hearing or communicating efficiently. The work done in this doctoral study aims at providing workers with good hearing protection and communication simultaneously without hindering the effectiveness of one for the other. The best option currently on the market is level dependent HPDs equipped with a boom microphone and radio capabilities. However, the weaknesses with such devices are two fold: the captured speech signal is degraded by the background noise and the radio does not distinguish intended listeners: all information is transmitted to everyone on the same radio channel. Therefore, the objectives of this work is to provide a high quality communication signal transmitted to an "intended" set of listeners within a specific spatial range; this range is defined by the talker’s vocal effort and the background noise level. An intra-aural HPD containing an in-ear microphone, a miniature loudspeaker, and an outer-ear microphone, with wireless radio capabilities and a link to a digital signal processor is used to achieve this task. Since it is intra-aural, when worn correctly, this HPD can provide a good passive protection from the noise. Also, when it is worn correctly, it creates an acoustical seal in the ear canal causing the phenomenon known as the occlusion effect. This occlusion effect causes an amplification of bone conduction sounds to the occluded ear, thus allowing for a speech signal to be captured under the earplug using the in-ear microphone. Since it is captured after the passive attenuation of the HPD, this speech signal has a relatively high signal-to-noise ratio. However, originating from bone and tissue conduction it has a limited frequency bandwidth. In this work, the speech signal captured with the in-ear microphones is enhanced by removing any residual noise and by artificially extending its frequency bandwidth. A normalized least mean squares adaptive filter, with a voice activity detection algorithm is used to denoise the signal, while the nonlinear effects of cubing the excitation signal are exploited to extend the bandwidth of the in-ear microphone speech. In this work, the enhancement algorithm is validated both subjectively and objectively. To mimic a natural acoustical environment by transmitting communication only to a set of intended listeners, the talker-to-listener distance is modeled as a function of the background noise level and variations in vocal effort. A direct relationship between these three factors is found after an experiment with 12 participants. Changes in speech levels and fundamental frequency as a function of noise and talker-to-listener distance for talkers wearing HPDs is determined. This knowledge paves the way for a radio acoustical environment, where a natural acoustical environment is mimicked while wearing HPDs and speech is only transmitted to a set of intended listeners within a given spatial range. The objectives of this work are met by providing a high quality speech signal only to an intended set of listeners. The contributions of this work include several peer-reviewed publications, conference proceedings, a provisional patent, as well as the development of functional earplug prototypes that could significantly enhance the communication experience of people wearing HPDs and thus increase the use of HPDs in noisy industrial environments and reduce the risk of noise induced hearing loss.

Date	8 Jun 2016
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Jérémie Voix (Supervisor) & Tiago H. Falk (Co-supervisor)