Distribution of short-term rms levels in conversational speech

For 30 male and 30 female talkers, distributions of short-term rms speech levels, relative to the corresponding long-term rms levels, were determined in each of eight 1/3-oct bands for six short-term measurement intervals. Consecutive, Hanning-windowed, 20-ms time records were combined to produce nominal measurement intervals ranging from 20 to 120 ms. For each measurement interval, mean distributions of short-term rms speech levels relative to long-term levels were very similar for male and female talkers, and intertalker differences were small, especially for short-term amplitudes above the median level. The distribution of short-term rms speech levels relative to long-term rms speech levels varied with measurement interval for the short-term measurements. The effects of measurement interval was least for the highest amplitude speech levels and increased as speech levels decreased. For short-term amplitudes above the median level, the effect of measurement interval was greater in higher frequency regions, whereas for short-term amplitudes below the median level, measurement interval had the greatest effect on the lower frequency bands. These data may facilitate comparisons among investigations using different measurement intervals. In addition, they have implications for amplification strategies.