Preliminary evaluation of a formant enhancement algorithm on the perception of speech in noise for normally hearing listeners.

Joseph I. Alcantara,G. Dooley,P. Blamey,Peter M. Seligman
DOI: https://doi.org/10.3109/00206099409072951
AUDIOLOGY
Abstract:The effects on speech perception in noise of dynamic filtering with bandpass filters centred at the first formant (f1) and second formant (f2) frequencies were evaluated with four normally hearing listeners. Multitalker babble was added to the speech signal with signal-to-noise ratios of -5 to -15 dB, chosen to reduce intelligibility to about 50%. The combined signal was then filtered with two-pole programmable bandpass filters centred at f1 and f2 under the control of a real-time speech processor. The f1 and f2 frequencies were estimated from the speech signal before noise was added to avoid hardware processing errors. Closed set vowel and consonant tests (using 11 /h/vowel/d/ and 12 /a/consonant/a/ stimuli), the Consonant-Nucleus-Consonant Monosyllabic Word Test and the Bamford-Kowal-Bench Sentence Test were carried out for three filter bandwidths (3/4, 1/3 and 1/6 octave) and for unprocessed speech in noise. The processing produced a small significant improvement for vowels in all three processed speech conditions and for monosyllables at the broadest filter setting compared to the unprocessed speech condition. There was no significant effect on consonants. A small negative effect was observed for sentences at the narrowest filter setting.
What problem does this paper attempt to address?