Harmonic Speech Coding - Digital Speech: Coding for Low Bit Rate Communication Systems - page 257

8
Harmonic Speech Coding
8.1 Introduction
A general sinusoidal analysis and synthesis concept was introduced by
McAulay [1] when he developed the Sinusoidal Transform Coder (STC) [2]
to demonstrate the applicability of the technique in low bit-rate speech
coding. Sinusoidal coding does not restrict the component sinusoids of the
synthesized speech to be harmonics of the fundamental frequency. The
frequency tracks of the sinusoids may vary independently of each other.
However in harmonic coding the higher frequency sinusoids are restricted
to be integer multiples of the fundamental frequency [3]. Therefore har-
monic coding can be seen as a subset of a generalized sinusoidal trans-
form coding. At low bit-rates, STC also restricts the frequency tracks to
be harmonics of the fundamental frequency, and deduces the harmonic
phases at the decoder, simply because the available bits are not suffi-
cient to encode the large number of parameters of the general sinusoidal
representation.
The STC was introduced as an alternative to the source filter model, and
its analysis and synthesis was directly applied to the original speech signal.
The binary voicing decision of the source filter model is one of its major
limitations. The STC employs a more general mixed-voicing scheme by
separating the speech spectrum into voiced and unvoiced components, using
a voicing transition frequency abovewhich the spectrumis declaredunvoiced.
However, one of the most recent harmonic coders operates in the LPC
residual domain, i.e. Split Band LPC (SB-LPC) [4]. SB-LPC replaces the binary
excitation of the source-filter model with amore general mixed excitation, and
filters the excitation signal using an LPC filter. The LPC residual has a simpler
phase spectrum than the original speech. The residual harmonic phases