Unvoiced sounds

Marques and Almeida, 1988] Marques, J. and Almeida, L. (1988). Sinusoidal modeling of speech Representation of unvoiced sounds with narrowband basis functions. 

The source characteristics of consonants differ depending on the class of sound being produced. All unvoiced sounds use only the noise source nasals and approximants use only the periodic source while voiced obstruents (i.e. voice fricatives, affricates and stops) use both sources. Approximants are generated in much the same way as vowels. Some consonants (such as [h]) can be synthesised in the same way as vowels that is by sending a sound source through the oral cavity resonators. Most other consonants are however more complicated than /h/ because their source is not at one end of the vocal tract. 

The first important category of sounds are known as oral stops. In English, stops occur in voiced/unvoiced pairs, so that [b] and [p] use the same manner and place and of articulation, and differ only as to whether the glottis is vibrating or not. The three nasals (the last sounds in THEM, THIN and thing) are often also described as nasal stops. While it is possible to continuously utter these three phones, they usually follow the pattern of the stop in normal speech, with a closure and... 

When we consider one particular pattern, the analysis seems clear pet and bet are different words, so /p/ and / must be different phonemes. It is reasonable to ask though, how we know that the soimd at the start of pet is the same sound as at the start of say pit There is no known deterministic or objective way to prove that this is the case, all we can do is rely on judgment and state that in terms of articulation, the two sounds are both produced with the lips, are unvoiced are stops and so on, and that if we examine them acoustically we see they are similar. But they are rarely so similar to be taken as identical their articulation and acoustic patterns are affected by the following vowel and this does have some effect. To a certain extent, there is a leap of faith required to say that pet and pit start with the same phoneme. This problem is even more acute when we consider sound contrasts in different positions ... 

If we consider the words span, stan and scan, we see that we have a three way contrast, in which the second sound differs in terms of place of articulation. There is however no voieing contrast possible at these points, so there are only three possible phonemes, unlike in word initial position where we have three place of articulation contrasts in addition to a voicing contrast, giving us six phonemes in all. The question then is, should span be transcribed as /s p ae n/ or /s b ae n/ The phonetic evidence tends to favour /p/, but this is not conclusive, and will vary from speaker to speaker and accent to accent. Clearly though there is a systematic effect going on. There is probably no objective answer to this, so we obey the normal convention (which may have been infiuenced by the spelling) and represent these words with the unvoiced phonemes /p/, 1x1 and Dd. 

Figures 12.4 and 12.5 show some typical power spectra for voiced and unvoiced speech. For the voiced sounds, we can clearly see the harmonics as the series of spikes. They are evenly spaced, and the fundamental frequency of the the source, the glottis, can be estimated by taking the reciprocal of the distance between the harmonics. In addition, the formants can clearly be seen as the more general peaks in the spectrum. The general shape of the spectrum ignoring the harmonics is often referred to as the envelope.

Given the durations, we ean ealculate the source signal for the utterance. We use an impulse source for sonorant phones, a noise source for unvoiced consonants, and a combined impulse and noise souree for voiced obstruents. The source characteristies are switched at phone boundaries. For voiced sounds, the impulse sequence is created by plaeing impulses at a separation distanee determined by 1 /FO at that point. Finally we feed the souree signal into the filter coefficients to generate the final speech waveform for the sentence. 

What then distinguishes the unvoiced fricatives such as [f] and [s] The answer lies in where the constriction occurs. For [f] the teeth combine with the hps to form the constriction this is termed labiodental. In [s] the sound is generated by the tongue... 

Figure 11.16 LTI filter models for various types of sound production (a) vowel and approximant, (b) nasalised vowel, (c) nasal, (d) unvoiced obstruent and (e) voiced obstruent.

---