KJBob

"Voice Stress Analyzer" for SigmaDSP

Discussion created by KJBob on May 2, 2013
Latest reply on May 3, 2013 by BrettG

    It's in quotes because no claims are made for accuracy, merchantability, or fitness for a particular purpose.  Rather, this is a fun gadget with which to impress your friends and kill time.  As the theory goes, the human voice normally features "microtremors" -- tiny, rapid variations in pitch.  When a person is thinking excessively about what they're saying (perhaps because they're busy making up a whopper), their microtremors may decrease.  Given something to measure, an engineer is likely to build something to measure it -- and SigmaDSP makes this seductively easy.

 

VSA_Schematic.jpg

 

    The circuit begins with gain/filtering/limiting, then a frequency demodulator to detect the microtremors.  The demodulator runs phase-detector feedback around a VCO to produce the inverse function of frequency-to-voltage.

 

    The frequency demodulator's output is filtered for the typical microtremor rate of five to 20 Hz.  This signal's amplitude envelope forms the final output.  A track-hold freezes it during pauses, to keep the display coherent.  Tests with a made-up input shows full-scale on the graph for a ~50Hz peak frequency deviation, with ~12 Hz showing half-scale.  This nonlinear response is from the square-root block, which expands the lower range for readability.

 

    The attached .dspproj file runs directly on a ADAU1701MINIZ eval board, and could be easily modified for others.  Gains are included to work with a cheap hi-Z dynamic mic plugged directly into the board's input jack without a preamp (the resulting lousy S/N doesn't seem to hurt).  Omit the gains (both discrete and inside the input filter) to use a preamp-level input.  The project's only output is the Real-Time Display.  While speaking, adjust the volume slider to find the point where the display begins to update, then set it a few dB higher.

 

    Experimenting with the voices of various persons on TV gives fascinating results. Typically, when the subject is speaking of more or less prepared lofty thoughts, the display trends equally lofty.  On the other hand, their microtremors often diminish while fielding tough questions.  Beyond this I can't say -- try it and see.  With work, this thing might even serve a useful purpose, such as analyzing a singer's vibrato.  Just please don't use it to judge any serious matter -- or antimatter, or doesn't matter!

Attachments

Outcomes