Applied Speech and Audio Processing: With matlab examples
Download 2.66 Mb. Pdf ko'rish
|
Applied Speech and Audio Processing With MATLAB Examples ( PDFDrive )
- Bu sahifa navigatsiya:
- Speech communications
5.2. Parameterisation
97 Figure 5.6 Parameterisation of the speech signal into components based loosely upon the human speech production system. In the sections that follow, many speech coding techniques will be introduced and discussed – but all related back to this parameterisation of the speech signal into different components that relate in some way to how they were produced. 5.2.1 Linear prediction Linear prediction has, for several decades, been the mainstay of speech communications technology, and applied to speech coding since at least 1971 [1]. It relies upon several characteristics of speech derived from the fact that speech is produced by a human muscular system. These muscles act to shape speech sounds through their movement, which is limited by a maximum speed. Muscles cannot move infinitely fast and thus, although human speech varies considerably throughout an utterance, it actually remains pseudo-stationary for around 30 ms (explained further in Section 2.5.1). In actual fact, the action of the glottis to generate pitch spikes is often shorter than 30 ms, so through some clever processing (see Section 5.3.2) pitch needs to be removed from the speech signal first – leaving a much lower energy signal called the residual. Pseudo-stationarity implies that the 240 samples at 8 kHz sample rate (corresponding to a duration of 30 ms), being similar, can be parameterised by a smaller set of values: typically eight or 10 linear prediction coefficients. Linear prediction coefficients are 98 Speech communications generator polynomials for a digital filter that, when stimulated with some input signal, recreates the characteristics of the original samples. They may not appear identical in the time domain, but most importantly their frequency response will match the original. Linear predictive coding (LPC) has been used successfully, by itself, in speech coding: the very low bitrate US Federal Standard 1015 2.4 kbits/s algorithm, developed in 1975, is based on LPC. It does, however, have very low perceived quality, and therefore limited to use in military communications. More normally LPC is refined with some other techniques discussed in this section when used in more modern coding algorithms. In this section we will cover the LPC filter itself, and then look at how to extract LPC parameters from the original speech signal residual. Some other important features relating to LPC stability and quantisation will also be discussed. Download 2.66 Mb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling