716: Final project, assignment 1 due by class time Tuesday 11/19

Data files: wav files and C/V segmented TextGrids

• In the Dropbox folder 716_creak/data, with one speaker in each subdirectory
• Seoyoung: hmong: Hmong wav and TextGrid files that include the "m" and "s" tones as target tones
• Claudia: cant: Cantonese wav and TextGrid files that include the "21" and "22" tones as target tones
• Alex: beijing: Beijing Mandarin wav and TextGrid files that include "21" and "51" tones as target tones
  • KY: needs to delete non 21/51 tone files
    

Hash files: tells you what tones were uttered for which item

• In the Dropbox folder 716_creak/info

Task 1: Run VoiceSauce on all files for your language

1. For each speaker, using Praat, determine an appropriate f0 range to set and record this (write it down in a file called "language"_f0_ranges.txt", e.g., hmong_f0_ranges.txt, or if you do a spreadsheet, hmong_f0_ranges.xls, etc.
   • Beijing: LOW: T3 (21), T4 (51). HIGH: 55 (T1) /35 (T2)
   • Cantonese: LOW: T4 (21), T6 (22). HIGH: T1 (55), T2 (25)
   • Hmong: LOW: m (21), s (22). HIGH: b (55), g/j: 42/52
2. The F0 ranges file should be a spreadsheet with the following columns: speaker, low_f0, high_f0, file_low, file_high, floor_f0, ceiling_f0. The low_f0 and high_f0 are the actual min/max f0 values you observed for the speaker. file_low and file_high are the item/file, e.g., 35_a you observed the value in. The floor_f0 and ceiling_f0 values will include an added buffer to use for VoiceSauce settings. for instance, suppose for a speaker you get high_f0 = 256 Hz. You might add an additional buffer of 15 Hz so celing_f0 = 256+15 = 271 Hz.
3. Note: we will just keep the f0 floor value at 40 Hz to allow for creaky items, but it's still interesting to note the low f0s you measured.
4. Run one speaker directory at a time in VoiceSauce, making sure to set the f0 floor and ceiling values individually for each speaker according to your results from Step 1 for the ceiling f0 value. For the f0 floor value, just use 40 Hz.
5. Make sure you save the generated Matlab files .mat files in the same directory as the wav and TextGrid files
6. The parameters to be measured are (under Parameter Estimation > Parameter Selection:
   • F0 (Straight)
   • F0 (Praat)
   • Formants (Praat)
   • Everything else below that, except last thing (epoch, excitation strenght)
7. Make sure you have the following settings:
   • F0: Used for parameter estimation: Straight
   • F0: Set min F0 for Straight and Praat to 40, and max f0 to your ceiling f0 value for the speaker
   • Formants and Bandwidths: Used for parameter estimation: PRaat

Task 2: Check over VoiceSauce files

1. Use the visual display in VoiceSauce to look over the VoiceSauce output
2. Check that the formant values are reasonable and report on this (write this up).
3. Make notes of anything else that looks anomalous.

Task 3: Generate an output to text file for your lanaguage (with all speakers in one file)

1. Click on Output to Text
2. For the Input .mat directory choose the parent directory for your language, e.g., beijing. Then click Include subdirectories and it will include your speaker-specific sub-directories Choose to Include TextGrid labels and Use subsegments, No. of subsegments 9
3. Name the output file "language"_cv_9subseg.txt", e.g., beijing_cv_9subseg.txt
4. Write the output file to 716_creak/analysis/