search

CSTR-Edinburgh / merlin

This is now the official location of the Merlin project.
http://www.cstr.ed.ac.uk/projects/merlin/
Apache License 2.0
1.31k stars 440 forks source link

Can't get merlin_synthesis.sh to work #444

Closed kwoot closed 5 years ago

kwoot commented 5 years ago

Hi, I am making a podcast to compare different tts programs on Linux. I can generate the full voices using run_full_voice.sh but generating a wav after that with my own simple text string seems undoable. How difficult is it, once the learning is done, to generate a new wav file from a string? Kind regards, Jeroen Baten

ZackHodari commented 5 years ago

You need to generate utterance structures using a TTS front-end, these can be converted into label files (.lab).

This script should get you started: https://github.com/CSTR-Edinburgh/merlin/blob/master/egs/fls_blizzard2017/s1/scripts/prepare_labels_from_txt.sh

See this for more context on usage: https://github.com/CSTR-Edinburgh/merlin/blob/master/egs/fls_blizzard2017/s1/merlin_synthesis.sh

RasmusD commented 5 years ago

That's not technically true. You need to generate label files. The utt files are an intermediate step that's not required. It just happens to be the easiest way to generate lab files using festival. But the links you provide should help somebody to generate label files consistent with those used for training assuming the user used the provided training scripts.