80 likes | 88 Views
The problem. 1.1 Background What is a voice for the brain? Source/filter theory of voice production: two independent components: larynx (f0) / vocal tract (F) How/where in the brain are these two independent cues processed?
E N D
The problem • 1.1 Background • What is a voice for the brain? • Source/filter theory of voice production: two independent components: larynx (f0) / vocal tract (F) • How/where in the brain are these two independent cues processed? • Recent advances in voice analysis/synthesis allow an independent manipulation of these two parameters • 1.2 Hypothesis • Manipulating the repetition of f0 vs. F in a series of voices should induce adaptation effects in different parts of auditory cortex. Pitch centre vs. temporal voice areas (TVA)? • Natural vs. unnatural f0/F pairings should lead to different activations in AC, in particular in the TVA • 1.3 Why is it important • Never been investigated; no idea based on what acoustic parameters is voice processed in the brain. Some data on f0, nothing on voice timbre. Better understand the network involved in voice processing • 1.4. Envisaged publication? • Not Science/Nature material, but PLOS/ cerebral cortex?
Experimental Design • Stimulation paradigm: • Pairs of voices uttering “had”: first voice unmodified (adaptor), second with modifications • 3 x 3 (x2) factorial design: • f0= -25%, 0, + 25% (to be piloted) • F= -15%, 0, +15% • Male/female voices • 18 types of pairs. X 20 different voices. 360 stimuli. SOA=4.1 sec. 25% null events as baseline. 33 min . • Task: forced-choice gender decision • Software: MCF (Media Control Function, Digivox, Montreal) • Hardware: Electrostatic headphones (NNL), response buttons • Volunteers: 20 normal volunteers with normal audition (self report ) • Post scanning behavioral data: naturalness & genderness ratings of each stimulus. • MRI parameters: • Rapid event-related design, ‘history controlled’ + ‘efficiency optimized’ order • Independent Voice localizer (10-min optimized bloc design) • Sequences : EPI. Moco (PACE?) • Spatial coverage: small number of slices to cover auditory cortex while allowing a 1-sec silent gap between volumes (to be optimized) • Timing: TR 2s, 270 volumes per run, 4 runs, overall duration: 42 min EPI + 8 min anatomical. About 1 hr15 per participant.
Analysis • Analysis package: BVX • Voxelwise approach (auditory cortex) + Functional localizer: Temporal Voice Area as a functional ROI • Fixed (within subject)+ Random effects design • Pre-processing details: standard pipeline • Analysis strategy: General Linear Model • Level of confidence: new to BV
Expected Results • Expected results, e.g.: • f0 and F : 240 vs. 120 stimuli per subject • Different regions of auditory cortex involved; overlap in TVA? • Natural vs. unnatural pairings: either 80 vs. 80 images per subject, AND/OR regression with naturalness scores • Regression with gender • Strategy, e.g.: • a) optimal result : adaptation to f0 vs F in single pairs • b) fallback options or additional cool result (higher risk): effect of naturalness
Summary of requested CCNi resources • 5.1 Stimulation • Auditory stimulation (NNL) • 5.2 Response • Response pad • 5.3 Number of scanning hours • 20 x 1.30 = 30h • 5.4 Analysis tools • BVX • 5.5 GRID use? • No (until BV is available on the grid) • 5.6 Storage space • Raw : 20 Gb • Analyse ??
Source Larynx, vocal folds => glottal pulses Filter Vocal tract => formants Voice production Fitch. (2000) Trends Cogn Sci
1,1 F 0,9 f0 0,8 1,2
NATURALNESS GENDER MALE VOICES f0 F F f0 FEMALE VOICES F f0 F f0