vinay ppt (2)
TRANSCRIPT
-
8/8/2019 VINAY PPT (2)
1/26
AN ANALOG INTEGRATED-CIRCUIT VOCAL TRACT
PRESENTED BY:VINAY VENUGOPAL
NO 66
S7 E&C
GUIDED BY:Ms. NIMMY GEORGE
LECTURER
E&C
-
8/8/2019 VINAY PPT (2)
2/26
2
OUTLINE Introduction Speech production Speech Locked Loop Circuit model of vocal tract Two port section Modeling of impedances Driving the vocal tract Conclusion
-
8/8/2019 VINAY PPT (2)
3/26
3
INTRODUCTION First experimental Integrated-circuit vocal tract.
Bio-inspired model
Analysis-by-Synthesis
-
8/8/2019 VINAY PPT (2)
4/26
4
SPEECH PRODUCTION
-
8/8/2019 VINAY PPT (2)
5/26
5
Contd.. Vocal tract is a cavity in animals where the sound that is
produced is filtered
Consists of laryngeal cavity, pharynx, oral cavity, nasalcavity. Lungs act as power supply Larynx modulates the airflow from the lungs
Vocal tract spectrally shapes the source
-
8/8/2019 VINAY PPT (2)
6/26
6
Contd.. Speech production is classified into 3 general
categories
1. Periodic2. Noisy3. Impulsive
Walls of the vocal tract controls the spectrum of thespeech radiated at the lips
-
8/8/2019 VINAY PPT (2)
7/26
7
SPEECH LOCKED LOOP
-
8/8/2019 VINAY PPT (2)
8/26
8
Contd.. Analysis-by synthesis method Speech is analyzed and parameters are extracted from it
to configure a speech synthesizer SLL is similar to PLL Measure of error is computed SLL locks to the input signal with the optimum vocal
tract profile
-
8/8/2019 VINAY PPT (2)
9/26
9
CIRCUIT MODEL OF
VOCAL TRACT Vocal tract can be approximated as a non uniform
acoustic tube with time varying cross-sectional areas
The cross sectional area is varied by varying theimpedance at different points along the tube
-
8/8/2019 VINAY PPT (2)
10/26
10
Contd.. Wave equation for 1d propagation of sound in a
uniform tube of circular cross section is
P-Sound pressureU-Volume velocity
Propagation of sound is accompanied by energy lossesdue to viscous friction and heat conduction by walls
-
8/8/2019 VINAY PPT (2)
11/26
11
Contd.. Acoustic propagation is analogous to plane wave
propagation through electrical transmission line
-
8/8/2019 VINAY PPT (2)
12/26
12
Contd..
Schematic dig of transmission line vocal tract
-
8/8/2019 VINAY PPT (2)
13/26
13
Contd.. VT is represented as acoustic tubes (intra oral and oral
tract) using transmission line model (TL)
Concatenation of many acoustic tubes Each 2 port is a LC circuit element Current source is used as volume velocity source at the
glottis
Current source is implemented using a Wide LinearRange Operational Transconductance Amplifier
-
8/8/2019 VINAY PPT (2)
14/26
14
This is converted to Tunable 2 port sections bygyrating RC network using WLROTA
Passive circuit model assuming rigid walls
-
8/8/2019 VINAY PPT (2)
15/26
15
SNR is calculated to be 64,66,67 Db
ID-V
DSof typical N MOS for various gate voltages
-
8/8/2019 VINAY PPT (2)
16/26
16
MODELLING OF
IMPEDANCES Glottal constriction resistance Zgc is implemented as
series of linear and nonlinear resistance
Implemented with MOS transistor Gate potential must be biased at the point given by
intersection of MOS device curve and desired I-V chara Linear Chara IV
Non-linear CharaIV
-
8/8/2019 VINAY PPT (2)
17/26
17
CONTD..
Gm is varied by varying Igm
-
8/8/2019 VINAY PPT (2)
18/26
18
DRIVING THE VOCAL
TRACT The area function space has large no of degrees of
freedom
To reduce dimensionality we use Maeda articulatorymodel
-
8/8/2019 VINAY PPT (2)
19/26
19
Contd.. Maeda model describes the vocal tract profile using
seven component
1. Jaw height2. Tongue body position3. Tongue body shape4. Tongue tip5. Lip height6. Lip protrusion7. Larynx height
-
8/8/2019 VINAY PPT (2)
20/26
20
Contd.. Articulatory codebook contains mapping from
articulatory and acoustic domains
Babble is produced using set of vocal tract profiles They are compiled into a look up table to produce
codebook- Babbling
-
8/8/2019 VINAY PPT (2)
21/26
21
Contd..
-
8/8/2019 VINAY PPT (2)
22/26
22
Contd.. Filter bank(130 to 6500) DCT( Discrete cosine transform) is applied to generate
a set of 12 Cepstral coefficients This is compared against the codebook Best match is found and corresponding articulatory
parameters are used to produce vocal tract area profile
-
8/8/2019 VINAY PPT (2)
23/26
23
CONCLUSION Low power Analog vocal tract chip can be used in SLL
to generate speech
Cross sectional area of tube can be varied by varyingL/C It can be used in speech synthesis, speech recognition,
compression etc
-
8/8/2019 VINAY PPT (2)
24/26
24
REFERENCES M. M. Sondhi and J. Schroeter, A hybrid time-frequency domain
articulatory speech synthesizer, IEEE Trans. Acoustics, SpeechSignal Process., vol. ASSP-35, no. 7, pp. 955967, Jul. 1987.
R. Sarpeshkar, M. W. Baker, C. D. Salthouse, J. Sit, L. Turicchia,and S. M. Zhak, An ultra-low-power programmable analog bionicear processor,IEEE Trans. Biomed. Eng., vol. 52, no. 4, pp. 711727, Apr.2005.
L. Turicchia and R. Sarpeshkar, A bio-inspired companding
strategy for spectral enhancement, IEEE Trans. Speech AudioProcess., vol.13, no. 2, pp. 243253, Mar. 2005.
-
8/8/2019 VINAY PPT (2)
25/26
THANK YOU
-
8/8/2019 VINAY PPT (2)
26/26
ANY QUESTIONS?