Analysis and Synthesis of Pathological Vowels

Size: px

Start display at page:

Download "Analysis and Synthesis of Pathological Vowels"

Earl Robinson
5 years ago
Views:

1 Analysis and Synthesis of Pathological Vowels Prospectus Brian C. Gabelman 6/13/23 1

2 OVERVIEW OF PRESENTATION I. Background II. Analysis of pathological voices III. Synthesis of pathological voices IV. Summary 2

3 BACKGROUND What is a pathological vowel? May be caused by physical or neural problems. Characterized by substantial and complex NONPERIODIC signal components. What is this project? Methods for modeling, analysis, and synthesis of pathological vowels incorporating novel approaches in: - System identification - Parameterization of non-periodic components (AM, FM, and noise) - Synthesizer designs for realtime and offline use 3

4 Why do it? BACKGROUND - Create a non-subjective basis to compare pathological voices for: 1. Improved diagnosis 2. Tracking changes in a patient s voice - Generate voice samples with known levels of variations (noise, roughness, etc.) for: 1. Evaluation of model parameters 2. Evaluation of listener variability 3. Evaluation of importance of levels of pathological features. What has been done before? - A well-established theory exists for NORMAL voices - Recent studies of pathological voices employ perturbation of normal features plus additive noise. - ES (external source) stimulation of the vocal tract to analyze formants (vowels) since 1942 for NORMAL voices. 4

5 BACKGROUND For the theoretical/analytical aspect of the project, an expression of the hypothesis of the dissertation in one sentence is: By means of FM and AM demodulation techniques, estimation of nonperiodic features of pathological vowels may be improved. 5

6 ANALYSIS SOURCE - FILTER MODEL OF SPEECH GLOTTAL SOURCE WAVEFORM (time domain) VOCAL TRACT FREQ. RESP. (freq domain) RESULTING VOICE SIG. (time domain) g(t) G(s) f(t) F(s) v(t) V(s) g(t) conv. f(t) = v(t) (time domain) G(s) x F(s) = V(s) (freq domain) 6

7 Steps for analysis: ANALYSIS PERIODIC ANALYSIS: 1. FORMANT DETERMINATION Uses LP (linear prediction) to model vocal tract as cascaded 2nd order digital resonators. External source testing is shown to augment or replace LP for pathological vowels (Inv). 2. SOURCE MODELING Uses inverse filtering and least squares optimization to fit source waveform to a standard model (LF). NONPERIODIC ANALYSIS: 3. ANALYSIS OF PITCH VARIATION Uses high resolution pitch tracking to measure detailed nonperiodic frequency variation. Variations are segmented into low and high frequeny FM with Gaussian form. 7

8 ANALYSIS 4. FM DEMODULATION Pitch variations are removed from original voice to achieve accurate noise estimation (step 7) (Inv.) 5. ANALYSIS OF POWER VARIATION Uses power tracking to measure detailed nonperiodic loudness variations. Variations are segmented into low and high frequency AM with Gaussian form. 6. AM DEMODULATION Power variations are removed from original voice to achieve accurate noise estimation (step 7) (Inv.) 7. ASPIRATION NOISE Frequency domain methods are used to separate aspiration noise component. The noise is spectrally modeled.[gus de Krom] 8

9 ANALYSIS BY SYNTHESIS ANALYSIS COLLECTION OF VOICE SAMPLES VOICE ANALYSIS + PARAMETERS (PITCH, NSR, FORMANTS,..) PERCEPTUAL COMPARISONS - VALIDATION VOICE SYNTHESIS 9

10 ANALYSIS ANALYSIS/SYNTHESIS MODEL OVERVIEW SOURCE WAVEFORM PERIODIC FM MODULATION SYNTHESIS ANALYSIS AM MODULATION NONPERIODIC + + VOCAL TRACT OUTPUT VOICE SPECTRAL SHAPING GAUSSIAN NOISE 1

11 ANALYSIS OVERVIEW OF PROJECT OPERATONS MIC SIGNAL LPC FORMANT ANALYSIS / MANUAL OPS PITCH TRACKER JITTER ESTIMATE. TREMOR TIME HIST POWER TRACKER SHIMMER ESTIMATE VOLUME TIME HIST FORMANTS PITCH TIME HIST POWER TIME HIST INVERSE FILTERING RESAMPLE TO REMOVE TREMOR MIKE SIGNAL RESAMPLE TO REMOVE TREMOR SYNTHESIZER RAW FLOW DERIVATIVE CONST. PITCH VOICE CONST. POWER VOICE LEAST SQR LF FIT CEPSTRAL NOISE ANALYSIS FITTED LF SOURCE PULSE SRC NOISE SPECTRUM NSR ESTIMATE 11

12 LINEAR PREDICTION PERIODIC ANALYSIS Estimates the vocal tract as an all-pole filter by minimizing the error between actual and model-predicted signals. (SOURCE) (VOCAL TRACT) (ERROR) u(n) 1 p a k z k = 1 G k s(n) (MODEL) + - e(n) UNKOWN SYSTEM H(z) PREDICTOR p α k = 1 k k z s (n) ESTIMATED INVERSE SYSTEM A(z) p k = 1 α G/H(z) k= 1 k z 12

13 PERIODIC ANALYSIS IDEALIZED LP RESULT Requires a priori knowledge of system. More difficult for pathological vowels. Imaginary part LPC COVARIANCE ROOTS. O=TRUE X=LPC POLE REFLECTED INSIDE UNIT CIRCLE Real part 25 LPC COV. PREDICTOR: LINE = ACTUAL OUT, DOT= PREDICTED ERROR SIGNAL OF COVARIANCE LPA = INVERSE FILTERED OUTPUT

14 PERIODIC ANALYSIS SOURCE-FILTER AMBIGUITY Source & filter are mixed in final voice. Unique LP solution may be difficult. CASE : NORMAL SOURCE AND NORMAL VOCAL TRACT SOURCE TIME SERIES VOCAL TRACT TRANSFER FUNCTION VOICE TIME SERIESVOICE SPECTRUM CASE 1: BREATHY SOURCE AND NORMAL VOCAL TRACT SOURCE TIME SERIES VOCAL TRACT TRANSFER FUNCTION VOICE TIME SERIESVOICE SPECTRUM CASE 2: NORMAL SOURCE AND ABNORMAL VOCAL TRACT SOURCE TIME SERIES VOCAL TRACT TRANSFER FUNCTION VOICE TIME SERIESVOICE SPECTRUM 14

15 PERIODIC ANALYSIS FITTING RAW SOURCE TO LF Having established the inverse filtered source, it is fit to the LF model [Qi] 15 E1 (FIRST SEG.) E2 (SECOND SEG.) FLOW U(t) & DERIVATIVE E(t) (tpk,epk) tp 3*E(t) U(t) (t2,ee/2) (te,ee) = E ( t ) dt tc T (SECONDS) E( t) = E 2B E ( t) = E E 2 1 ( t) = E ( t) = E e e e e αt e ε ( t t ) sinω t, ε ( t t e e g ), t e + m( t te), t t t e e t t c t t c 15

16 NON-PERIODIC ANALYSIS HIGH RESOLUTION PITCH TRACKING Nonperiodic analysis begins with interpolating pitch tracking. 8 6 VOICE SIGNAL 4 VOICE SIG Measured Period TIME (SEC) PITCH TRACK FREQ (HZ) TIME (SEC) 16

17 NON-PERIODIC ANALYSIS FM DEVIATION SEGREATED TO LOW AND HIGH FREQUENCY The pitch track is low/hi pass filtered to yield tremor and HFPV (High Frequency Pitch Variation) FREQ HZ SEC 4 2 FREQ HZ SEC 17

18 NON-PERIODIC ANALYSIS GAUSSIAN HFPV Successful pitch tracking yields a Gaussian distribution in HPFV. The standard deviation is a convenient measure of HFPV #OCCURRANCES DELT FREQ (%) TOTSKIP= FCUT=1 TOTMAN= 18

19 NON-PERIODIC ANALYSIS FM DEMODULATION The pitch track may be used to demodulate the original voice to obtain a version with almost no pitch variation; re-tracking verifies constant pitch (<.1%)..1.5 DELTA FREQ PERCENT TIME (SEC) 19

20 SMOOTHED ABS ORIG VOICE x PITCH TRACK FEATURES SAMPLE NUMBER NON-PERIODIC ANALYSIS POWER TRACKING Analogously to pitch tracking, voice power is tracked. ENVELOPE MINIMA POWER TRACK.8 POWER x 1 4 SAMPLE # 2

21 POWER SEGREGATED TO LOW AND HIGH FREQUENCY NON-PERIODIC ANALYSIS The power track is low/hi pass filtered to yield low frequency power variations and high frequency shimmer. 2 x 1 8 ORIGINAL POWER ( ) AND TREMOR (LINE) 1.5 POWER TIME (SEC) 2 SHIM% = 1*(POWER - LOWPASS POWER)/POWER DELTA POWER PERCENT TIME (SEC) 21

22 NON-PERIODIC ANALYSIS GAUSSIAN POWER VARIATIONS Shimmer also displays Gaussian variations. 4 SHIM% HIST. STAND DEV = 4.823% #OCCURRANCES DELT PWR (%) 22

23 NON-PERIODIC ANALYSIS AM DEMODULATION The power track may be used to demodulate the original voice to obtain a version with almost no variation in strength; re-tracking verifies constant power. 1 VARIOUS MEASURES OF SIGNAL STRENGTH.95 POWER SUM ENERGY.9.85 MAX AMPL SAMPLE # x

24 LOG1 MAGNITUDE NON-PERIODIC ANALYSIS ASPIRATION NOISE ANALYSIS Aspiration noise is segregated via spectral techniques. Peaks in the FFT of the log of the FFT (cepstrum) represent periodic energy, and are filtered out with a comb filter (lifter). Results are used to calculate noise-to-signal ratio (NSR) LOG1 MAGNITUDE TIME (SEC) Figure 2.3c. Cepstrum (expanded scale) FREQUENCY 2 (HZ) TIME (SEC) Figure 2.3a. PSD oforiginal voice. Figure 2.3b. Cepstrum of original voice LOG1 MAGNITUDE TIME (SEC) Figure 2.3d. Comb-liftered cepstrum of 14c FREQUENCY (HZ) Figure 2.3e. Orig PSD, aspriation PSD, 1 FREQUENCY 2 3 (HZ) 4 5 and vocal tract. Figure 2.3f. Source aspiration PSD with vocal tract removed and fitted to 25-point piecewise-linear model. 24

25 NON-PERIODIC ANALYSIS FM DEMODULATION IMPROVES NSR ACCURACY Using FM demodulation improves resolution of spectral peaks of periodic components, thus allowing longer FFT windows and more accurate NSR determination LOG1 MAGNITUDE FREQUENCY (HZ) 25

26 NON-PERIODIC ANALYSIS CHANGES IN NSR AFTER FM AND AM DEMODULATION FM demodulation reduces NSR measures by up to 2 db, yielding results closer to perceived levels. AM demodulation has much less effect ORIG NSR (DB) TREMOR REMOVED ALL FM REMOVED CASE# - SORTED BY ASCENDING ORIGINAL NSR 26

27 PC #1: STIMULUS GEN. PERIODIC ANALYSIS EXTERNAL (ES) SOURCE ANALYSIS Source-filter ambiguity may be resolved by augmenting the glottal source with an known external stimulus. [Epps]. STIMULUS SIGNAL D/A AMPLIFIER XDUCER ACOUSTIC CONDUIT /a/ PC #2: DAQ MEM. CH 1 A/D CH 2 SIGNAL CONDITIONER MICROPHONE VOCAL TRACT A/D SIGNAL CONDITIONER TUBE OF LENGTH L WITH CLOSED END. FORMANTS AT F = C/4L, 3C/4L, 5C/4L,... 27

28 PERIODIC ANALYSIS ES VERIFICATION A simple plastic tube model verified the ES experimental setup. Resonances occur at expected frequencies SPECTRUM WITH RAG, W/O RAG, & MAG OF T.F. (YEL) HA(f) 1 HB(f) 8 MAGNITUDE (db) TUBE FORMANTS -2-4 V(f) FREQUENCY (Hz) x

29 ES: NORMAL /a/ LP & FFT analysis show consistent results with ES analysis for a normal vowel. PERIODIC ANALYSIS 3 2 SPECTRA D1: VC LPA(DASH), VC db, EXTSRC 2dB F1 F2 EXT SRC T.F. S F3 F4-3 LPA FFT

30 PERIODIC ANALYSIS ES: SIMULATED BREATHY /a/ LP & FFT analysis show poor resolution for F3 and F4 for a breathy /a/, while the ES resolution for F3 and F4 remains good. D1BRTHY: VOICE LPA & FFT (BOT), EXTSRC TF (TOP) F1 F2 EXT SRC T.F. S -1 MAGNITUDE (db) -2-3 F3 F4 LPA FFT FREQUENCY (Hz) 3

31 SYNTHESIS SYNTHESIS OF PATHOLOGICAL VOWELS Synthesis is a critical step in the study of pathological vowels. It provides evidence of the success of analysis and modeling steps via immediate comparisons of original and synthetic voice. Two synthesizers were implemented: 1. A realtime hardware-based synthesizer capable of providing instant response to changes in model parameters. 2. A software synthesizer implemented in MATLAB with extended features, convenient graphical interface, and ease of modification. 31

32 CURRENT REALTIME SYNTHESIZER FUNCTIONAL OVERVIEW SYNTHESIS REALTIME SYNTHESIZER - Implemented in native X86 assembly language - Executes all code within 1us cycle - Overrides PC OS to achieve determinancy - Employs dedicated clock, I/O, and control hardware implemented in a wire-wrap PCB adapter card IMPL KGLOT 88 SOURCE SELECTION + 2 POLES... 4 ZEROES AGC CLK D/A X86 INTERUPT LOW PASS FILT. LF1 GAH RECORD WAV AMP LF2 ARB ASP NOISE STORE/RECALL CONTROL PARMS OUTPUT SIGNAL FILE SPKR SOURCE CONTROLS ARBITRARY SOURCE FILE JITTER SHIMMER DIPLOPHONIA PARAMETER TIME VARIATION 32

33 SYNTHESIS VALIDATION SYNTHESIS The current model, analysis tools, and synthesizers yield a high level of fidelity in generation of synthetic pathological vowels. The system is currently employed at the UCLA Voicelab for NIH funded perceptual studies. In order to objectively validate the analysis/synthesis process, the loop is closed by re-analyzing the synthetic time series to confirm parameter values. Re-analysis also provides opportunity to observe interactions of nonperiodic components. 33

34 SYNTHESIS VALIDATION SYNTHESIS Re-measured synthetic aspiration noise level agrees with level set in synthesizer. -5 MEASURED NSR IN SYNTHETIC (DB) A.N. SET NSR IN SYNTHESIZER NPB21 34

35 SYNTHESIS SYNTHESIS VALIDATION Aspiration noise adds about %.2 to HFPV measurements. JITTER MEASURED IN SYNTHETIC (%) JITTER LEVEL SET IN SYNTHESIZER (%) HFPV adds about 4dB to NSR measurements. MEASURED NSR IN SYNTH (db) A.N. SET NSR IN SYNTHESIZER (db) 35

36 SYNTHESIS SYNTHESIS VALIDATION Subjective analysis by synthesis experiments demonstrate the success of AM and FM demodulation in achieving accurate modeling of nonperiodic features. Listeners adjust synthetic aspiration noise to match original. Match improves with demodulation -5 ORIGINAL VOICE (NO DEMOD) PEARSON =.51-1 SABS ASPIRATION NOISE (DB) CEPSTRAL NSR (DB) 36

37 SYNTHESIS VALIDATION SYNTHESIS SABS ASPIRATION NOISE (DB) TREMOR REMOVED PEARSON = CEPSTRAL NSR (DB) SABS ASPIRATION NOISE (DB) ALL AM&FM REMOVED PEARSON = CEPSTRAL NSR (DB) 37

38 SUMMARY CONCLUSION This study has achieved improved automatic, objective analysis and synthesis of speech within the specialization of pathological vowels. Specific accomplishments include: - A unique, symmetric model for nonperiodic components as AM, FM and spectrally-shaped aspiration noise - Improved accuracy of noise analysis via AM & FM demodulation - Application of ES formant identification for pathological vowels. - Implementation of realtime and offline specialized high fidelity vowel synthesizers 38

Synthesis Algorithms and Validation

Chapter 5 Synthesis Algorithms and Validation An essential step in the study of pathological voices is re-synthesis; clear and immediate evidence of the success and accuracy of modeling efforts is provided