Speech referenced dynamic compression limiting: improving loudness comfort and acoustic safety

This thesis presents the research, development and evaluation associated with a novel method of improving listening comfort and acoustic safety for people listening primarily to speech produced by electronic devices such as headsets, telephones, headphones, hearing aids, cochlear implants, level-dependent hearing protectors and public address systems. The method involves a novel technique of amplitude limiting audio signals that convey speech. Time-varying, frequency-specific levels of speech generate a set of time-varying speech reference levels. The method limits the level of the audio to these speech reference levels and hence is called speech referenced limiting (SRL). In principle, SRL provides the greatest limiting of noise for the least limiting of speech, making it arguably the optimal method for limiting noise in speech systems. Two schemes based on the method were developed, the SRL MKI and the SRL MKII schemes. The latter scheme was far superior, with the ability to estimate the speech loudness and power from frequency regions where speech was dominant, while ignoring frequency regions where it was not. It contained a novel method of determining the amount of additional control needed to correct for the loudness summation of noises with a bandwidth exceeding that of speech, as well as providing fast speech referenced control over the power of abrupt sounds while introducing only a very short delay.Subjective evaluation of the SRL MKI and SRL MKII schemes conducted in the laboratory confirmed large reductions in noise loudness and preservation of speech quality. It was hypothesised that the SRL MKII scheme would provide the greatest reduction in the excess loudness of an audio signal compared with the loudness of the preceding speech conveyed by the audio signal for the least reduction in the speech loudness and quality. Using stimuli typical of those experienced in the three main intended applications (hearing aids, level-dependent hearing protectors and telephone headsets), this hypothesis held true and noise control was shown to be far superior to a conventional fixed-reference limiter while speech loudness and quality were maintained.