The signal-description has the following format:
- comments are started by a '#' as first char of line
- syllable-borders are denoted by a single '-' as first char of line
       followed by an integer denoting the accent-type: 0 for none,
       1 for word-accent, 2 for sentence-accent
- feature-flags have the syntax: <+|-> <flag> [time of occurance in msec]
- possible feature-flags are:
    +/- voc, brt, nas, lar, spr for voiced, breathy, nasalized, laryngealized, spread respectively
    these features can be valid in regions and are followed by a time value
    +rounded, +overshoot, +undershoot for the corresponding vowel qualities
    these features are valid for the whole phone
- phone descriptors follow the MBROLA-syntax:  <name of phone> <dur in msec> <intonation description>

here's an example:

_ 50
a 75 0 106 6 106 12 106 18 106 24 103 30 103 36 98 ...
-lar 30 50
n 45 0 97 11 103 22 103 33 103 44 103 55 102 66 111 77 111 88 111
- 0
d 10 0 111 50 111
e: 65 0 111 7 111 14 108 21 107 28 108 35 108 42 ...
n 45 0 103 11 99 22 95 33 95 44 95 55 95 66 95 77 95 88 95
- 0
v 35 0 95 14 95 28 95 42 95 56 95 70 95 84 99
o: 55 0 99 9 99 18 99 27 103 36 107 45 111 54 111 ...
- 2
x 50
@ 30 0 178 16 175 32 174 48 168 64 165 80 160
n 45 0 153 11 156 22 153 33 151 44 149 55 143 66 134 77 132 88 127
- 0
...

This format can be generated automatically from the MBROLA-format by use of a syllableborder marking program.

To distinguish it from the mbrola-format, it's refered as eph-format (for extended pho) in the documentation