2
$\begingroup$

I've read that the reflection coefficients in speech processing (as computed by the Levinson-Durbin algorithm for solving the Yule-Walker equations) "represent the fraction of energy reflected back" at each tube junction,[1] assuming the speaker's vocal tract is modeled as a series of uniform lossless acoustic tubes (see Figure 1). Now, I've seen that statement from multiple PDFs online, but did not see it mentioned in my textbook (Rabiner, Schafer - Digital Processing of Speech Signals).

Does this mean we can do autoregressive modeling of electrical transmission lines as well? In other words, does a series of lossless transmission lines form an all-pole filter whose (magnitude) reflection coefficients (one for each junction) and linear-prediction coefficients can be estimated by inputting white noise to the system and then analyzing the output with the Levinson-Durbin algorithm? (Figure 2.)

Figure 1. Approximating a time-varying acoustic tube using a series of short tubes of uniform cross-section

Figure 1. Approximating a non-uniform acoustic tube using a series of short tubes of uniform cross-section. (Source: Rabiner, Schafer - Digital Processing of Speech Signals p.83.)

Figure 2. A series of lossless transmission lines.

Figure 2. A series of lossless transmission lines. (Source: Pozar - Microwave Engineering 4th Ed. p.251.)

[1] http://research.cs.tamu.edu/prism/lectures/sp/l7.pdf

$\endgroup$
1
$\begingroup$

This is a neat idea and could probably be made to work - it seems like a variant on time-domain reflectometry, which exploits correlations of this kind https://en.wikipedia.org/wiki/Time-domain_reflectometry. However, there's an important difference between your two use cases: we can normally inject a test signal onto an electrical transmission line, while we normally can't do that in the vocal tract. Measuring the cross-correlation of the injected and reflected signals is generally going to be more reliable than measuring the cross-correlation of noise.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.