Shunsuke
NAKATA
1. Three levels of analysis
At the study of speech communication, three
levels of analysis can be distinguished, that is, linguistic, perceptual and
acoustic ones. It means that the research of speech sound can be made possible
only within the limits of acoustical and perceptual experiments.
2. Linguistic level
1)
Data of informants Date of informants, such as their age ,sex,
origin, profession can be obtained by questionnaires. Generally, to raise the
statistical reliability of the data, the number of the informants should be
augmented.
2)
Type of speech materials To analyse the communicative value of
intonation, the 都pontaneous conversation will best be used, but other
authentic documents such as radio or TV recordings have often been preferred in
order to reduce the complexity of discourse structure. Researches which treat
exclusively the so-called neutral intonation to analyse its syntactic function
restrain their materials to texts or sentences read out.
3)
Number and volume of speech materials
It is generally recommended to prepare plural corpora of almost the same
volume. The number and the volume of speech materials should be determined
according to the purpose of the analysis.
3. Perceptual level
Intonation can be described as complex
patterns of three acoustical parameters, i.e. F0 , intensity and
duration. In modelling intonation, several pitch levels are preliminally
determined, which have to be validated by perceptual experiments. Actually, it
is confirmed that not all the pitch movements are linguistically pertinent.
4. Acoustical level
Digital analysis of speech sound has
structural limits as below. It need to be controled according to the
caracteristics of materials and to the the purpose of the investigation.
Extraction
of F0 : Trade-off between time and frequency resolution
Measurement
of duration : Difference between potential and effective resolution
Measurement
of intensity : Trade-off between time and amplitude resolution, S/N ratio