Praat TextGrid to MBROLA
synthesiser format (pho file) converter
Dafydd Gibbon, 2009-01-31
The seg format is:
<![CDATA[file name
| diphone name
| | first sample
| | | last sample
v | | | middle sample
V V V v
#a.d16 _ a 2000 3910 2929
ak.d16 a k 2000 4027 2939
i#.d16 i _ 2000 4654 3228
ki.d16 k i 2000 3882 3150]]>
Comments:
- The TextGrid should only contain 1 tier, with the items to be paired as diphones.
- It will be assumed that the associated WAV file is sampled at 16000 kHz.
If this is not the case, resample:
sox inputfile.wav -r 16000 outputfile.wav
- The phoneme notation in the Praat annotation should correspond to the phoneme notation in the MBROLA voice you intend to create.
- MBROLA requires silent segments to be annoted with '_' (underscore). However, in some environments Praat apparently does not permit the use of underscore, so silent segments should
- either be annotated with some other convention, and replaced by '_' in the TextGrid file before conversion,
- or be annoted with 'SIL', which will be automatically replaced by '_'.
- If you find errors, they may be due to special symbols in the labels. If you suspect this, please replace the special symbols, and let me know which symbols were involved.