LPC BASED SPEECH SYNTHESIS WITH ADAPTIVE PITCH PREFILTER

SYNTHETISEUR DE PAROLES A CODAGE PREDICTIF LINEAIRE DOTE D'UN PREFILTRE ADAPTATIF D'ACCENTUATION DE HAUTEUR DE SON

English Abstract

2066568 9106091 PCTABS00005
A linear predictive coding speech synthetizer wherein an
excitation signal that includes pitch information is filtered in a pitch
enhancement filter (201) prior to provision of the excitation
signal to an LPC filter (106). In one embodiment, the speech
synthesizer is configured in a radio.

Claims

WO 91/06091 PCT/US90/05191
Claims
1. A method of synthesizing speech, comprising:
A) providing an excitation signal that includes
pitch information; further characterized by the steps of:
B) filtering the excitation signal in a pitch
enhancement filter to provide a pitch filtered excitation
signal;
C) filtering the pitch filtered excitation signal in
a speech synthesis filter to provide a synthesized
speech signal.

WO 91/06091 PCT/US90/05191
2. A method of synthesizing speech, comprising:
A) receiving a coded information signal;
B) using the coded information signal to provide an
excitation signal that includes pitch information;
further characterized by the steps of:
C) filtering the excitation signal in a pitch
enhancement filter to provide a pitch filtered excitation
signal;
D) filtering the pitch filtered excitation signal in
a speech synthesis filter to provide a synthesized
speech signal.

WO 91/06091 PCT/US90/05191
3. A radio, comprising:
A) RF means for receiving a broadcast signal and
for recovering a coded information signal included
therewith;
B) excitation source means operably coupled to the
RF means for providing an excitation signal that
includes pitch information in response to the coded
information signal; wherein the radio is further
characterized by:
C) pitch enhancement filter means operably
coupled to the excitation source means for filtering the
pitch information in the excitation signal to provide a
filtered excitation signal;
D) LPC filter means for receiving the filtered
excitation signal and for providing a synthesized speech
signal in response thereto.

Description

~ wo gl/06091 ~ 2 ~ PC~/U59n/05191
LPC BASED SPEECH SYNTHESIS WITH ADAPTIVE PITC~
PREFILTER
10 T~chnical Field
This invention r~lates generally to speech
synthesis, and more particularly to linear predictive
coding based speech synth~sis.
1~ .
a~ round ~f thQ Inv~ntiQn
The synthesis of sp~ech through use of a linear
predictiv~ coding (LPC) based platform is known in the
20 art. A prior art radio that embodies such a platform is
depicted generally in Fig. 1 by the r~ference numeral
100. The radio (100) receives a spoech coded signal
(101) through an appropriate energy transducer (102),
such as an antenna. An RF unit (103) converts the
25 received ~ignal (101) to basaband and demodulates the
signal to recover the spsech ooded information.
A parameter dscoder (1û53 dcvclops control parameters
for various subsequen~ proc0ss~s from this informa~ion.
An ~xcitation sourc~ (104~ as describ~d above utilizes
30 the parameters provided to it to creatc an ~xcitation
signal and then provides that excitation signal (which
excitation signal includes pitch information that has
been inserted by a pitch filter) to an LPC filter (106)
C~l lR!~ TI lTF ~,~

Wc~ 9~/06091 ` `` ` ; 2 a ~ 8 P~/US90/Oil9(1~
which in turn provides, at it's output, a synthesized
speech si~nal.
This synthesiz~d speech signal is then filtered in
an adaptiv~ pitch postfilter (107) and an adaptive
5 spectral postfilter (108), as well as a post emphasis
filter (109), to enhance the percep~ion ef natural speech
and to minimiz~ the impact of various distortions and
artifacts introduced in the synth0sis process. The
enhanced synthesiæed speech signal is then properly
10 processed in an audio processing unit (111) and rendered
audible through an appropriate audio transducer (112).
Each of the above generally rsferred to components
is well known and understood in the art.
The pitch postfilter (107) se~es an important
1~ function, in that it provides additional c~ntrol of the
pitch content of the synthesized speech. Without this
filter, the resultant syn~hesized speech product may be
rougher and of low~r quality. Notwithstanding this
important banefit, the pitch postfiiter (107) frequently
20 contributes artifacts to the r~sultant synthesized
speech, which artifacts can themselves noticeably
disturb th~ perc0ption of natural spe~ch. Aecordingly, a
need exists for providing appropriat0 pitch enhancement
filtcring in ar~ LPC based speech synthesizing unit ~hat
25 minimizes a conourr3nt peroaptible expression of
a~ifacts in a r~sultant synthesiz~d sp~ech signal.
This need and o~hers are substantially met through
provision of the LPC plafform disclosed herein.
According to the inv~ntion, the pitch enhancement
!SUE~STITUTE St~

,~` WO 91/D6091 . 2 0 ~ ~ a 6 8 P~/IIS90/!)5191
postfilter is moved from a position of processing the
LPC filter output to a position where it processes the
~xcitation input to the LPC filter. The subsequent
proc~ssing of the LPC filter its~lf functions to minimize
5 the perceptible effect of any artifacts introduced by the
pitch enhancement pr~filter in th~ resultant synthesized
speech signal.
~rief l~escrip~iQ~ of ~ rawin
Fig. 1 comprises a block diagram dopiction of a
prior art embodiment of a radio receiver having LPC
based speech synthesis; and
Fig. 2 compris~s a block diagram depiction of a
15 radio having LPC bas~d speech synthesis in aGcordancs
with the invention.
In Fig. 2, a radio embodying the invention includes
an antenna (102) for receiving a spcech coded signal
(101). An RF uni~ (103) processes the received signal to
recover the spe~ch coded information. This information
is provided tu a parameter decoder (105) ~hat develops
control parameters for various subsequent processes.
An excitation source (104) as described above utili2es
the parameters provided to it to create an excitation
signal. This rasultant exeitation signal from the
exoitatiGn source ~104~ is provid~d to a pitch prefilter
(201 ) that functions to filt~r the pitch information
oon~ained in the excitation si~nal. The resultant filtered
signal then passes to the LPC filt~r (106) which yields a
8UBilal~JTE S~EET
'

WO 91/06091 4 ~ 0 6 ~ 5 S 8 ~'1`/US90/~191~,~
synthesized speech signal in accordance with the coded
information.
Consistent with known methodology, this resultant
signal is then further processed in an adaptive spectral
5 postfilter (108) and post emphasis filter (109) to
furth~r ~nhance the quality of the synthesi~ed speech,
and is then processad in an audio processing unit (111)
and rendered audible by an audio transducar (112).
What is claimed is:
~;IL9 B~ ITU7~E 51~3EET

Representative Drawing