Documentation on the structure of the ETL Formant data file:
		"MokhtariTanaka2000_ETLformantdata.txt"

The formant data file contains 8 columns and 2750 rows of 
formant frequency and bandwidth values represented in floating-
point format in units of Hz, to the nearest 0.1 Hz (i.e., to
1 decimal place).


The 8 columns are structured as follows:

F1	F2	F3	F4	B1	B2	B3	B4

where {F1,...,F4} and {B1,...,B4} are the first four formant
frequencies and bandwidths, respectively.


The structure of the 2750 rows is represented by the following,
nested-loops:

Speakers 1,2,...,5
  Vowels 1,2,...,5
    Words 1,2,...,22
      Frames 1,2,...5

where the Speaker-codes in the original "ETL-WD-I and II"
dataset are:
	Speaker 1 --> S0001
	Speaker 2 --> S0003
	Speaker 3 --> S0010
	Speaker 4 --> S0015
	Speaker 5 --> S0041

and the vowels are:
	Vowel 1 --> i
	Vowel 2 --> e
	Vowel 3 --> a
	Vowel 4 --> o
	Vowel 5 --> u

Note: The phonetic symbol of Japanese "u" looks like a
joined double-u "/uu/" (please see the full paper).

The list of 22 words for each of the 5 vowel-nuclei can be
found in the full paper.

The 5 frames are consecutive frames at the steady-state of
each vowel nucleus (see full paper for details).

---