[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

*To*: AUDITORY@xxxxxxxxxxxxxxx*Subject*: Re: [Fwd: technical notes on data used by Martin Braun]*From*: Paul Boersma <paul.boersma@xxxxxxxxxx>*Date*: Mon, 11 Jun 2001 19:09:17 +0200*Delivery-date*: Mon Jun 11 13:04:02 2001*In-reply-to*: <3B24C582.7587D6F7@psychologie.uni-leipzig.de>*References*: <3B24C582.7587D6F7@psychologie.uni-leipzig.de>*Reply-to*: Paul Boersma <paul.boersma@xxxxxxxxxx>*Sender*: AUDITORY Research in Auditory Perception <AUDITORY@xxxxxxxxxxxxxxx>

>> Was the pitch contour presented as a continuous line, or was it >> quantized in quarter semitones? > >Well, it was displayed as a series of frame-by-frame values, not as a >continuous line. At the 16k sampling frequency that we used, the >frame-by-frame values have a resolution of a quarter of a semitone. Is it likely that the measured F0 values had a spacing of 1/4 semitone? It seems more likely that they were expressed as an entire number of samples per period. For instance, for an F0 of 200 Hz you would get 16k/200 = 80 samples per period. In that vicinity, then, the spacing is 1 sample per period, or 1.25%, or just below one quarter of a semitone, which seems to be consistent with what Bob Ladd remembers. However, if you "bin" such values into 1/4-semitone buckets, you should control for the number of possible values that fit into such a bucket. In the range 115-230 Hz, that number is 1, 2, or 3. Of course, I don't expect any researcher to have overseen such a correction, but it is interesting to see what results such mistake would have led to. The following is a Praat script that generates random F0 values between 115 and 230 Hz, and bins them according to the above faulty criterion: for bin from 10 to 57 n'bin' = 0 endfor for i to 10000 f0 = randomUniform (115, 230) samplesPerPeriod = round (16000 / f0) f0_gipos = 16000 / samplesPerPeriod semitones = hertzToSemitones (f0_gipos) quarterSemitoneBin = round (semitones * 4) n'quarterSemitoneBin' = n'quarterSemitoneBin' + 1 endfor echo Bin n for bin from 10 to 57 n = n'bin' printline 'bin' 'n' endfor In the bin range from 10 to 57 (that is, 48 bins = one octave), we now find distinctive peaks at bin numbers 25, 27, 29, 31, 33, 35, 38, 41, 45, and 52. Note that this series contains several subseries at distances of 4, 6, or 8 bins (4 bins = 1 semitone). Can the authors show us how they corrected for this sampling-and-binning problem? Does a reanalysis with a program that has an F0 measurement accuracy of, say, 0.00001 Hz instead of 1/4 semitone, still lead to the reported result? Best wishes, Paul -- Paul Boersma Institute of Phonetic Sciences, University of Amsterdam Herengracht 338, 1016CG Amsterdam, The Netherlands http://www.fon.hum.uva.nl/paul/ phone +31-20-5252385

**Follow-Ups**:**Re: [Fwd: technical notes on data used by Martin Braun]***From:*Alain de Cheveigne'

**Re: [Fwd: technical notes on data used by Martin Braun]***From:*Martin Braun

**References**:**[Fwd: technical notes on data used by Martin Braun]***From:*Christian Kaernbach

- Prev by Date:
**[Fwd: technical notes on data used by Martin Braun]** - Next by Date:
**Input Device** - Previous by thread:
**[Fwd: technical notes on data used by Martin Braun]** - Next by thread:
**Re: [Fwd: technical notes on data used by Martin Braun]** - Index(es):