1. AN ANALOG INTEGRATED- CIRCUIT VOCAL TRACT PRESENTED BY: NIEL V JOSEPH S7 AEI ROLL NO-46 GUIDED BY: MR.SANTHOSHKUMAR.S ASST.PROFESSOR E&C DEPARTMENT

2. CONTENTS  Introduction  Human vocal tract  Concept of speech locked loop  Circuit model of vocal tract  Two -port ∏ -section  Linear and non linear resistor modeling  Driving the vocal tract  conclusion 2

3. INTRODUCTION  First experimental integrated circuit vocal tract  16 stage cascade of two port ∏-elements  Analysis by synthesis  Speech locked loop 3

4. Human vocal tract 4

5.  Function is filtering of sound  Consist of laryngeal cavity, pharynx, nasal cavity and oral cavity  Length in adult males is 16.9 cm and in females 14.1cm  Larynx produces sound in mammals  Lungs act as power supply 5

6.  controlled variations in the area of vocal tract produces speech  Two sources of excitation are  Periodic source at the glottis  Turbulent noise source at some point along the tube  Vocal fold vibrations produces interruption of airflow 6

7. CONCEPT OF SPEECH LOCKED LOOP FIG: CONCEPT OF SPEECH LOCKED LOOP 7

8.  Analysis- by -synthesis  Analogy with phase locked loop  Measure of error is computed by comparing to the input  SLL locks to the input sound with optimal vocal tract profile  Vocal tract is analogous to VCO in PLL 8

9. CIRCUIT MODEL OF VOCAL TRACT  Vocal tract is assumed as non-uniform acoustic tube  Terminated by the vocal chord at one end and lip/nose at other end  Cross sectional area is varied by changing impedances at different points  Propagation of wave is approximately one dimensional 9

10.  The wave equation for one dimensional sound propagation in a uniform tube of circular cross section is P-sound pressure U-volume velocity A-area of cross section C-velocity of sound 10

11.  Acoustic wave propagation in a tube is analogous to plane wave propagation along an electrical transmission line  Equation can be modified as V-voltage I-current 11

12. FIG: SCHEMATIC DIAGRAM OF TRANSMISSION LINE VOCAL TRACT 12

13.  Transmission Line (TL) model  TL comprises of cascade of two-port elements  Current source model  Variable impedance model  Fluid volume velocity is mapped to current  Fluid pressure is mapped to voltage 13

14. TWO-PORT ∏-SECTION FIG: PASSIVE ∏ CIRCUIT MODEL ASUMING RIGID WALLS 14

15. FIG: CIRCUIT DIAGRAM OF TUNABLE TWO PORT ∏ -SECTION 15

16. FIG:MEASURED SIGNAL AND NOISE CHARARACTERISTICS AS A FUNCTION OF INPUT FREQUENCY  SNR IS 64,66,63 dB for F1,F2,F3 of /e/ 16

17. LINEAR AND NON LINEAR RESISTOR MODELING  Implemented with MOS transistor  Glottal constriction resistance is a series combination of linear and non linear resistors  For linear characteristics I ∞V  For non linear characteristics I ∞√V 17

18. FIG: I -V CURVES FOR TYPICAL nMOS TRANSISTOR 18

19. FIG: MEASURED I-V CHARACTERISTICS OF MOS RESISTOR CONFIGURED AS A 100 GIGA OHM RESISTANCE 19

20. DRIVING THE VOCAL TRACT  It can produce all speech sounds  We should be given area function, the glottal excitation source, the turbulent noise source  Area function has large number of degrees of freedom  To reduce the dimensionality we use Maeda articulatory model 20

21.  The Maeda articulatory model describes the vocal tract profile using seven components 1.Jaw height 2.Tongue body position 3.Tongue body shape 4.Tongue tip 5.Lip height 6.Lip protrusion 7.Larynx height 21

22.  Articulatory code book contain mappings from the articulatory to acoustic domain  Set of vocal tract profiles are generated  ‘babble’ is synthesized using each vocal tract profile 22

23. 23

24.  DCT is applied to generate a set of 12 cepstral coefficient  Compared against the codebook  Best match is found and corresponding articulatory parameters are used to produce vocal tract area profile 24

25. CONCLUSION  Consumes less than 275 micro watt power  Used in speech locked loop to generate speech  Cross sectional area of tube can be varied by varying L/C  Also used in speech compression an speech recognition 25

26.  For many speech synthesis applications 5-7kHz is sufficient Spectrogram of recording of the word ‘Massachusetts’ Synthesized by AVT, female vocal tract is used 26

27. REFERENCES 27  B. Raj, L. Turicchia, B. Schmidt-Nielsen, and R. Sarpeshkar, “An FFTbased companding front end for noise-robust automatic speech recognition,” EURASIP J. Audio, Speech, Music Process., vol. 2007, 2007, 10.1155/2007/65420, Article ID 65420.  R. Sarpeshkar, M. W. Baker, C. D. Salthouse, J. Sit, L. Turicchia, and S. M. Zhak, “An ultra-low-power programmable analog bionic ear processor,” IEEE Trans. Biomed. Eng., vol. 52, no. 4, pp. 711–727, Apr. 2005.  L. Turicchia and R. Sarpeshkar, “A bio-inspired companding strategy for spectral enhancement,” IEEE Trans. Speech Audio Process., vol. 13, no. 2, pp. 243–253, Mar. 2005.

28. 28

29. 29