icvgip 2012 icvgip 2012 speech training aids visual feedback of the articulatory efforts during...
TRANSCRIPT
ICVGIP2012
ICVGIP2012Speech training aids
Visual feedback of the articulatory efforts during acquisition of speech production by a hearing-impaired child.
Display of articulatory effort using LPC-based analysis of speech signal
• Oral cavity: fixed length tubular sections.
• LPC analysis of windowed speech frames
>> LPC reflection coefficients >> Section area ratios >> Section areas, assuming constant glottis-end area >> Vocal tract shape [Wakita, 1973]
>> Display of the articulatory efforts not visible on speaker's face.
Introduction
ICVGIP2012
ICVGIP2012
Problem: Errors due to variation in glottis-end area during speech production [Wakita,1979] .
Proposed solution•Acquisition of speech as audio and facial image as video.•Using mouth opening area estimated from the video as the reference area of the lip-end section, for scaling of the area ratios obtained from LPC analysis of simultaneously acquired speech signal [Nayak et al., 2012] .
Investigation A technique for estimation of the mouth opening, without errors caused by teeth and tongue between the lips• Contrast enhancement with multi-threshold binarization • Connected component detection
ICVGIP2012
ICVGIP2012
Processing steps
iv) Horizontal opening
v) Vertical opening: segmentation, multi-threshold
binarization, connected component detection
vi) Det. of inner lip boundaries vii) Mouth opening area calculation
i) Input frame ii) Face sub-image iii) Mouth sub-image [Viola & Jones, 2004] [Hsu et al., 2002]
ICVGIP2012
ICVGIP2012
Test resultsTest results
•Test material: video recordings of vowels /a i u/ of 12 male speakers.
•Scatter plot of estimated values & values obtained manually
•Corr. coeffi.: 0.91