"for "pau"" features in HTS labels

[Etienne D]

Hello,
I am currently going through all the input labels, working on the temporary files created by utts_to_mlf.sh that serve as a basis for the awk scripts that convert it to monophones and full-context labels:
# oo th 0 0 1 0 0 1 0 . . . 0 2 0 7 0 5 # @ 1.0700001 1.258

I’m using the scheme script label.feats as a reference for the meaning of each feature, however there is a series of features (49 to 62) that I do not seem to understand:
the script simply says “for “pau””, and each feature seems to give information about whether the previous/next syllable/word/phrase is stressed/accented, or give its length/part-of-speech…
Some of the features seem to have almost the same values as previous (but identically named) features, but not quite. They Additionally, the pauses in HTS labels are marked ‘#’, not ‘pau’ (whether that matters or not).
I have searched in quite a few of the scripts and the HTS website, but could not find relevant information.
Could you please clarify what these “for “pau”” features stand for?

Thank you.

[Etienne D]

Hi,

I went as far as possible with many other things, but I still need to know about those label features if I am to build a full front-end.
Does anyone have any further information about how the values of these features are defined?

Thanks.

[Simon]

These features are apparently especially for short pauses (we think) but their origin is somewhat lost in time. I suspect you may be able to omit them with only a little degradation (try it and see for yourself though).

[Author]

Posts