Singing synthesis frontend for espeak
Switch branches/tags
Nothing to show
Clone or download
divVerent UST parsing fixes for me rindo eng.ust:
- Ignore non-INI-conforming "UST Version1.2" line.
- Intensity is not always defined. Prefer Velocity, and if both are
  missing, default to 100.
- Apparently, in this file the tempo is stored in MIDI form (µs per
  quarter note), not in bpm. Add a workaround until we know more.
- Better error handling.
Latest commit 749c6b1 Jun 1, 2017
Permalink
Failed to load latest commit information.
eCantorix split off common functions Oct 27, 2012
examples Redo ASS support. Oct 19, 2016
voiceshape
.gitignore Fun feature: stdout now outputs lyrics karaoke subtitles in .ass format. Jul 18, 2013
COPYING GPL3 Oct 16, 2012
README.rst fix lyrics track detection for older abc2midi Aug 10, 2013
TODO add TODO Jul 18, 2013
ecantorix.pl
ust2mid.pl
vsq2mid.pl

README.rst

eCantorix

(*)       (*)
 |         |
 |--=====--|
 |  |||||  |
 |  |||||  |
 ;  |||||  :
(   |||||   )
 \  |||||  /
  `-=====-'

eCantorix is a singing synthesis frontend for espeak. It works by using espeak to generate raw speech samples, then adjusting their pitch and length and finally creating a LMMS project file referencing the samples in sync to the input file.

Dependencies

  • espeak
  • sox
  • LMMS
  • Perl * Config::Tiny * MIDI::Opus * Math::FFT

How to use

The input file for eCantorix is a slightly more strict MIDI karaoke (.kar) file.

Lyrics have to be stored as text events in the same track as the notes associated with them.

A lyrics event then covers all notes starting between itself and the next text event.

You can use an empty text event to end the range of notes covered by lyrics.

In case the range of a lyrics event covers multiple notes, the notes must be all connected (legato), as they will be performed as a single audio sample with pitch shifting. This feature however yields poor quality lyrics and is not recommended - it tends to sound better to rather use separate lyrics events and thus split a syllable into two, like "soft" -> "saw-oft".

For polyphony, you can either use multiple tracks (with each having their own instance of the lyrics), or a single track and have the notes placed on multiple MIDI channels (they then will share the lyrics).

Once you have such an input file, run:

perl ecantorix.pl -O mmp -o foo.mmp foo.mid

The result is a LMMS project file that can be processed to an audio file using:

lmms -o foo.wav --render foo.mmp

You can also render to a wave file directly:

perl ecantorix.pl -O wav -o foo.wav foo.mid

Also, there is now preliminary support for Vocaloid (VSQ) files. The phonemes are not used yet, though, as doing this would require an external mapping table that depends on the Vocaloid voice in use.

License

This software, as well as espeak, is licensed under the GPL version 3 (see the included file COPYING) or any later version, at your choice.

The author believes that the license of the program does not affect licensing or copyright of the output files; therefore you can use the output files for any purpose of your liking.

The example files are licensed as follows:

examples/entchen.abc
This is a wellknown German children's tune. Public domain (unknown author, but sure dead for more than 70 years).
examples/nacht-v.abc
This is from the Zauberflöte (Magic Flute). Public domain (Wolfgang Amadeus Mozart).
examples/sadi-moma.abc
This is the Free Software Song. No copyright is claimed on this song or its ABC typesetting by me, Richard Stallman, or the FSF.
examples/sadi-moma-pitchbend.abc
This is the Free Software Song. No copyright is claimed on this song or its ABC typesetting by me, Richard Stallman, or the FSF.
examples/sadi-moma-qon.abc
This is the Free Software Song. No copyright is claimed on this song or its ABC typesetting by me, Richard Stallman, or the FSF.
examples/sarastro.abc
This is from the Zauberflöte (Magic Flute). Public domain (Wolfgang Amadeus Mozart).