The hearing aid input: a phonemic approach to assessing the spectral distribution of speech

The long-term goal is to investigate the feasibility of using real speech as a stimulus for electroacoustic evaluation of nonlinear hearing aids. The goals of the present study were to determine the spectral envelope of speech from acoustic measures of phoneme tokens in running speech, to compare the results with published data on long-term average speech spectra, to measure inter-talker differences of spectral envelope, and to explore the extent to which the intensity variation within and across talkers might be minimized by frequency-selective amplification and automatic gain control.

Seven phonemes were selected to represent the extremes of frequency and intensity in English. Recordings were made of five men and five women producing syllable strings constructed from these phonemes. One-third octave spectra were prepared from the phoneme tokens. The frequencies and intensities of 13 key points in these spectra were measured and used to estimate individual and group spectral envelopes.

The group spectral envelope was similar to that derived from published data on the long-term average spectrum of speech, but there were marked intertalker differences. Some of the differences were gender-related. The overall dynamic range of intensity in these data was 53 dB. Frequency-dependent level adjustment (an 11 dB high-frequency boost) reduced this range to 42 dB, and a combination of frequency-dependent and subject-dependent level adjustment (analogous to 2-band automatic gain control) reduced it to 37 dB.

A phonemic approach to determining the spectral envelope of speech offers insights that are not available from long-term average spectra and could offer advantages in the evaluation of nonlinear hearing aids.