Applied Speech and Audio Processing: With matlab examples
Characteristics of speech
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
3.2. Characteristics of speech
43 Table 3.1. Amplitude of speech in several environments, from [9]. a Location Noise level (dB SPL ) Speech level (dB SPL ) school 50 71 home (outside, urban) 61 65 home (outside, suburban) 48 55 home (inside, urban) 48 57 home (inside, suburban) 41 55 department store 54 58 on a train 74 66 in an aircraft 79 68 a These data were originally published in The Handbook of Hearing and the Effects of Noise, K. Kryter, Chapter 1, Copyright Elsevier (Academic Press) 1994. signal-to-noise level results (i.e. where the speech level is lower than that of the noise), and that is partly due to the particular shape of the noise spectrum in both environments. Later, in Chapter 6 we will derive a way of taking account of differences in speech and noise spectra. In general, as the noise level increases by 1 dB, a speaker will raise his voice level by 0.5 dB within the range of normal speech [9]. With very low noise levels, a male adult speaker can produce 52 dB SPL of speech measured at a distance of 1 m when speaking casually. This raises to about 90 dB SPL when shouting. Quoted figures of 50–82 dB SPL for women and 53–83 dB SPL for children were probably obtained by someone with no experience of child raising. The dynamic range of conversational speech is around 30 dB [10], and the mean level for males measured at 1 m is somewhere in the region of 55–60 dBA SPL 2 with peaks that extend 12 dB beyond this [6]. 3.2.3 Types of speech Speech can be classified into phonemes which fall into several different types (see Infobox 3.1: The structure of speech on page 40), whereas the phonemes themselves are best identified using the international phonetic alphabet (see Infobox 3.2: The Inter- national Phonetic Alphabet on page 41). When spoken, phonemes will naturally have different average amplitudes. These are shown classified by type in Table 3.2, where the relative amplitude of spoken phonemes in each class is listed, along with the amplitude range of phonemes within that class. More detail, including phoneme-by-phoneme examples from the original experiments, can be found in [11]. 2 The ‘A’ in dBA refers to the A-weighting curve (discussed in Section 4.2.1), in which a frequency correction is made to the measured signal prior to obtaining the average amplitude. This frequency correction is based upon ‘hearing curves’ and attempts to make the measured value more representative of what a human would perceive. |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling