Trevor Martin voice recording

Hello -

I did some detailed additive analyses of stretches of single voiced melody recorded from students singing notes at Florida State University. The recordings of their sung notes were digitally recorded with series of voltage amplitudes as measured by the microphone at a rate of 40,000 words per second recorded in the form of data sets.

I analyzed this recorded sound data into collections of partials, their amplitudes and frequencies being computed at 0.5 or 1.0 millisecond intervals (one point on an amplitude or frequency trajectory every 20 or 40 points of the 40 KHz voltage vs time plot from the A/D converting microphone).

For each stretch of sung melody I analyzed, one of the parameters I computed was an "audio frequency spectrum" for a stretch of an individual's singing. Although the pitch (frequency) of the tone being sung might change by as much as a few semitones (as in a musical melody) over the stretch of sound, the audio spectrum I computed for that stretch of sound would hold fairly constant for a sound stretch being sung by the same individual making the same vowel sound and changing little but the pitch of the sound.

I worked backwards to resynthesize voicelike musical tones from the analysis data - audio frequency spectra, pitch, sound amplitude, etc. I had derived for stretches of sung sound for various singers who had contributed sung notes for recording (and then A/D conversion and analysis using a state of the art (~1991) musical analysis/synthesis I had the use of a Florida State University.

I worked with the data I had collected at FSU for years after collecting it. Observations I made which may be of interest here is that the audio SPECTRA I computed for an individual's stretch of melody had a "fine structure" - small kinks and elevations and depressions on that person's audio spectral plot which correlated with the sound of that individual's singing and vowel sound. Thus for a stretch of sung melody with a given vowel sound, an audio spectrum computed for it would be very characteristic for the individual close to the way a fingerprint would be.

I heard portions of recorded sound (played by NPR over internet radio) said to be recorded either from Trevor Martin or from Mr. Zimmerman during a 911 call. The voice sound communicated distress.

I am not set up to do an audio analysis of that sound now, but from listening to it, I'm persuaded it would be possible to COMPARE analyses of that 911 sound with: (A) other stretches of recorded sound made by Trevor Martin - hopefully some on hand from somewhere - and (B) stretches of recorded sound made by Mr. Zimmerman (which could be taken now if needed).

It strikes me that the recording of the 911 sound is plenty good enough to yield analysis results which would show quite clearly whether or not it was the sound of Mr. Martin's voice or the sound of Mr. Zimmerman's voice. I believe it could be determined by analysis of that sound that it was unlikely to be that of either Mr. Martin's or Mr. Zimmerman's voice if that was the case.

Even if no recordings of Mr. Martin's voice can be found, I believe it could be determined whether or not it was highly likely that the voice on the 911 recording was that of Mr. Zimmerman or not. It would be very desirable, however, to get hold of a recording of a stretch of Mr. Martin's voice, even if this were a brief stretch of not very high quality voice sound.

Reference: The Cro-Magnon Advanced Additive Analysis/Synthesis System by Ralph David Hill - Proceedings of the International Computer Music Conference 1991 Montreal.

Sincerely, Ralph David (Dave) Hill