Creating English gen .lab Files for Synthesis

0. Use demo labels

If you don't care about synthesizing specific sentences and just want some samples for how the voice sounds, you can use the "alice" generation labels that come with the demo and skip the rest of these instructions.

1. Put your utterances into the Festival .data format

( sus_00001 "sentence to synthesize." )
( sus_00002 "another new sentence." )
( sus_00003 "here is one more." )
....

2. Create Festival .utt files

$FESTVOXDIR/src/promptselect/text2utts -all -level Text -odir youroutputdirectory -otype utts -itype data "yourfile.data"

3. Convert Festival .utt format to HTS .lab format

Follow the steps in the data Makefile for make label. We typically just copy over the data/Makefile (and the data/scripts directory) to wherever the utts are, edit it to just do fullcontext, and run it.

Remove unnecessary timestamps

The training .lab files have timestamps for where the phonemes start and end, however these are not needed for the generation .lab files. Remove them. Speech lab students: see /proj/tts/examples/fix_labels.py for an example script that does this.