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

Re: VCV nonsense syllable corpus



Dear Abeer & List,

Your post below calls to mind work on micro-timing of music, where similar non-linearities were involved.  Henkjan Honing and colleagues took musical phrases from actual performances (which contain small “expressive variations” in timing and dynamics that are related to the rhythmic and melodic structure of the melody), and sped them up or slowed them down while preserving pitch and spectral information.  He found that the micro-timings did not linearly scale, but were characteristic of particular musical tempos.

I look forward to seeing Abeer’s data base information (and perhaps a few recent citations?  I have only been able to find older—but interesting—studies on VOT vs. speech rate in a hasty online search)

With thanks,
Justin London

On Oct 7, 2016, at 1:44 AM, ABEER ALWAN <alwan@xxxxxxxx> wrote:

sorry if you have received this message multiple times..i had some issues with the server. 


Hi,
 Variations in rate are not the same thing as shortening or lengthening an utterance (it is a non-linear process; the VOT, for example, varies differently than the vowel when rate changes and initial consonants vary differently than final ones, etc). The acoustic and perceptual consequences can be significant especially in difficult listening situations. We collected such a database a few years ago and I will try to sort it out and post it soon.

Regards,
Abeer


On Wednesday, October 5, 2016, Matthew Winn <mwinn2@xxxxxx> wrote:
Narayan,
If you or a colleague are familiar with Praat, you can use this Praat script to generate variations of your stimuli where there are manipulations of the pitch contour and the duration. You'll end up with your original sound, an versions where the pitch went up or down (three different pitch contours) and for each of those, a version with the original duration, as well as a shortened duration and lengthened duration. So 9 sounds total from each original sound you have. 
The variables should be clear from the top of the script. As is the case for any Praat manipulation, your new sounds will be temp objects, not saved files. You can save them with another script that does batch processing. But anyway, this kind of procedure might serve your needs. It works on a PC, not sure about a Mac. 

Matt

# Enter your sound directory here, within double quotes:
sound_dir$ = "C:\consonants"

# select your manipulations (these are multipliers of the pitch contour and duration)
pitch_up = 1.15
pitch_down = 0.85

duration_up = 1.2
duration_down = 0.8

# clear the info window
clearinfo

# list all the files in the folder
Create Strings as file list: "fileList", "'sound_dir$'\*.wav"
num_files = Get number of strings

# loop through the files
for index from 1 to num_files
 select Strings fileList
 filename$ = Get string: index
 name$ = "'filename$'" - ".wav"
 name$ = "'filename$'" - ".WAV"
 Read from file: "'sound_dir$'\'filename$'"

# make a manipulation object using some default pitch analysis parameters
 To Manipulation: 0.01, 75, 600

 # create various pitch tiers
   Extract pitch tier
   Copy: "'name$'_p2"
   Copy: "'name$'_p1"
   Multiply frequencies: 0, 1000, pitch_down
   select PitchTier 'name$'
   Copy: "'name$'_p3"
   Multiply frequencies: 0, 1000, pitch_up

 # create various duration tiers
   select Manipulation 'name$'
   Extract duration tier
   Copy: "'name$'_d2"
   Copy: "'name$'_d1"
   Add point: 0, duration_down
   select DurationTier 'name$'_d2
   Copy: "'name$'_d3"
   Add point: 0, duration_up

 for p_index from 1 to 3
   for d_index from 1 to 3

   select Manipulation 'name$'
   plus PitchTier 'name$'_p'p_index'
   Replace pitch tier
   
   select Manipulation 'name$'
   plus DurationTier 'name$'_d'd_index'
   Replace duration tier
      
   select Manipulation 'name$'
   Get resynthesis (overlap-add)
   Rename: "'name$'_p'p_index'_d'd_index'"
      
   endfor
 endfor

# cleanup
selectObject: "Sound 'name$'"
plusObject: "Manipulation 'name$'"
plusObject: "PitchTier 'name$'"
plusObject: "PitchTier 'name$'_p2"
plusObject: "PitchTier 'name$'_p1"
plusObject: "PitchTier 'name$'_p3"
plusObject: "DurationTier 'name$'"
plusObject: "DurationTier 'name$'_d2"
plusObject: "DurationTier 'name$'_d1"
plusObject: "DurationTier 'name$'_d3"
Remove

print finished with file: 'name$''newline$'
#pause finished with file: 'name$'

endfor


--------------------------------------------------------------
Matthew Winn, Au.D., Ph.D.
Assistant Professor
Dept. of Speech & Hearing Sciences
University of Washington

On Tue, Oct 4, 2016 at 9:55 AM, Narayan Sankaran <Narayan_Sankaran@xxxxxxxxxxx> wrote:

Hi all,

 

I’m looking for help in finding a corpus of VCV nonsense syllables. I’m examining the cortical representation of consonants (using EEG) and want capture some invariance to acoustic variability. More so than presenting multiple talkers, I am looking for multiple repeat utterances of the same VCV within the same talker (not only different vowel contexts, but the exact VCV uttered at different rates and with as much variability in prosodic features as possible). Does anyone on this list knows of such a corpus? Many thanks in advance.

 

Best,

Narayan

 

Narayan Sankaran 
Research Intern | Starkey Hearing Research Center 
2150 Shattuck Ave. | Suite 408 | Berkeley, CA 94704-1362

starkey.com | starkeyresearch.com | map | email

<image001.gif>

<image002.jpg><image003.jpg><image004.jpg>

 




*************************************************
Justin London
Andrew W. Mellon Professor of Music, Cognitive Science, and the Humanities
Carleton College
One North College St.
Northfield, MN 55057 USA
+1 507-222-4397