Abstract: The Long-term average spectrum of speech is important for both designing the system and rating the loudness in speech communication, and it deserves to be investigated in some detail.
The testing material is a testing sentence composed and used by the Institute of Acoustics "Ta qu Wuxishi, Wo dao Heilongjiang". Using the same sentence as testing material has the advantages to free the talker from looking at the written text and to keep the speech level constant easily.
There were 12 talkers (6 males and 6 females) whose daily speech is Standard colloquial Chinese in the testing crew. Every talker spoke repeately the testing sentence in an anechoic chamber at five speech levels measured at far-field (1 metre from mouth of talker):very soft (55 dB), soft (60 dB), normal (65 dB), loud (70 dB), very loud (75 dB), with normal utterance speed (about four syllables per second). At the normal speech level, talkers were instructed to speak the testing sentence with two speeds (normal and fast). The speech signals were analysed by a heterodyne analyzer (B & K 2010).
It's shown that the long-term average spectrum of speech depends obviously on the sound pressure level of speech. But little on utterence speed. The fundamental frequency is increased by about an octave and the spectrum level at frequency of 1400 Hz is increased by 30 dB, as the consequence of an increase of 20 dB in overall sound pressure level consequently, in the calculation of speech loudness, the long-term average spectrum must be considered as dependent somewhat on speech effort.
Both average values and standard deviations of the spectra for different over-all sound pressure level of speech were given.