Speech Data and Emu Templates

This page contains details of the data collected and analysed to produce a vowel space plot for AL.

Word List

The following list of words was used to record the monphthong vowels in an hVd context. Each word was recorded in isolation, using the Sound Recorder application under Windows XP. Each word was recorded four times. Phoneme boundaries were marked using EmuLabeller.

hod heard hid had hard hood heed whod head hudd

The following digit strings were also recorded using the Sound Recorder application, and word segmentation was performed using EmuLabeller. Each digit string was recorded twice.

9541 1249 3263 9874 6521 5169 6853 7268 5094 5320 4877 6687 5039 3300 2140 8790 1457 2121 3879 4806

Recordings of the above words and digit strings (sampled at 16bit, 11.025kHz, Mono) are available for download.

Also included in the same Speech Data download package are the Emu label files, which contain phoneme boundary markings and target indicators (for the hVd data), word segmentation (for the digit data), and information about the data source, such as speaker, gender, accent, data set, for both the hVd and digits data.

Emu Templates

To analyse the spectograms of the collected speech data, I used the EMU Speech Database System. The templates used to perform the labelling of phonemes, words, and other characteristics of the collected speech data are shown below.

Due to the lack of an automatic formant tracker, the formant values were extracted by eye, after the phoneme boundaries and targets had been marked using EmuLabeller, configured with the templates below.

The hVd Template

! template file for SLP801 data collection -- isolated words
! Andrew Lampert, June 2004

level Speaker
level Word       Speaker
level Phoneme    Word
level Target     Phoneme

label Speaker Sex
label Speaker Accent
label Speaker Set

labfile Phoneme :extension phn :time-factor 1000
labfile Target  :extension trg :type EVENT :time-factor 1000


! Here we define the legal label categories for various levels
! this gives us a menu in the labeller and we can use the
! category names in queries
!
legal Phoneme vowel A E I O V U ai ei oi i@ u@ au @u @: @ a: e: i: o: u:
legal Phoneme monophthong A E I O V U @: a: e: i: o: u:
legal Phoneme shortvowel A E I O V U
legal Phoneme longvowel @: a: e: i: o: u:
legal Phoneme diphthong ai ei oi i@ u@ au @u
legal Phoneme stop p tS dZ t k b d g
legal Phoneme nasal m n N M
legal Phoneme fricative f v s z S Z h D D- T
legal Phoneme approximant w j l r
legal Phoneme other H #

legal Target - T

legal Set - hVd digits
legal Sex - M F
! possible accent labels, you could add more if you don't fit one of these
legal Accent - Australian US Singapore Malaysia

! ************************************************************
! Define the location of files, you will need to modify this
! ************************************************************
path hlb,phn,trg,wav      D:\uni\801\Assignments\OwnVowelSpace\data

! this tells Emu to look for sampled speech data in wav files
track samples     wav

set PrimaryExtension wav


The digit Template

! template file for SLP801 data collection -- digit strings
! Andrew Lampert, June 2004


level Speaker
level Word       Speaker

label Speaker Sex
label Speaker Accent
label Speaker Set

labfile Word :extension wrd :time-factor 1000

legal Word - one two three four five six seven eight nine ten zero oh

legal Set - hVd digits
legal Sex - M F
! possible accent labels, you could add more if you don't fit one of these
legal Accent - Australian US Singapore Malaysia

! ************************************************************
! Define the location of files, you will need to modify this
! ************************************************************
path hlb,wrd,wav      D:\uni\801\Assignments\OwnVowelSpace\data\digits

! this tells Emu to look for sampled speech data in wav files
track samples     wav

set PrimaryExtension wav