System Documentation

Installation and Test

1. Download and unpack the distribution after registering as a new user.
2. Unzip the file. If your unzip tool has a "smart" option for unpacking text files, disable this - it can corrupt some of the data files in the code.
3. Run the file flinger_test.bat in the directory where you unpacked the distribution. This will run the program and load an example script called fl_test.scm. The script is written in Festival's scripting language, called "scheme". You should hear some talking and singing, synthesized from the MIDI file in festival\examples\midi\ogi.mid. You can look at this file in your favorite MIDI composition program.

Interactive Command Line Mode

• All of the commands in fl_test.scm could have been typed at the festival command line, such as
  festival> (Flinger.sing "festival/examples/midi/ogi.mid" nil)
  (Yes the parentheses are important!)
• To save the output to a wav file, do
  festival> (utt.save.wave (Flinger.sing "festival/examples/midi/ogi.mid" nil) "test.wav" 'riff)
  A file test.wav should be created in the standard Microsoft 'riff' format.
• To exit the program, type
  festival> (quit)
  and press Enter. (Yes the parentheses are important!)

  Scripts

  Festival and Flinger use a lisp-like scripting language called "scheme". The example script fl_test.scm will get you started. For more info, see the section on scheme in the Festival manual.

  Global Singer Parameters

  Global Singer Parameters are, in a sense, the properties of the "singer" who will be performing your MIDI file. These are static for the whole rendering of a MIDI file once you define them. In the test script fl_test.scm, you should find a code block that illustrates how to change these parameters, which are explained below:
  (Flinger.init
(list
'(vibrato_freq 5.0)
'(vibrato_max 0.030)
'(maxbend_semi 1.0)
'(portamento_time_default 0.050)
'(drift_freq1 4.7)
'(drift_freq2 7.1)
'(drift_freq3 12.7)
'(drift_ampl 0.003)
'(transpose -5)
'(consonant_stretch 1.0)
'(phone_delimiter "|")
'(dump_info 0)
))

  

  vibrato_freq

  frequency of vibrato

  vibrato_max

  maximum depth of vibrato (at max range of MIDI Modulation controller) as a percentage of the frequency of a note

  maxbend_semi

  frequency modification in semitones at max excursion of MIDI pitch bend controller

  portamento_time_default

  time of glide between notes (seconds)

  drift_freq1,2,3

  frequencies of quasi-random drift of frequency -- models the inability of humans to sing a steady note

  drift_ampl

  amplitude of drift, as percentage of note frequency

  transpose

  shift of note frequencies (in semitones) from those given in MIDI file

  consonant_stretch

  stretch of consonants in slowly sung syllables (experimental)

  phone_delimiter

  In the phoneme input mode (yet to be documented), the delimiter used to separate phoneme symbols.

  dump_info

  set to "1" to dump out lots of debugging information that may or may not be helpful

  MIDI-Controllable Parameters

  Using the MIDI controllers to dynamically modify the characteristics of the voice makes the difference between a very machine-like rendering of the melody and something pretty. The following MIDI controllers are used:
  • Modulation - controls vibrato depth
  • Pitch Bend - controls pitch bend depth
  • Volume (planned for future) - controls volume
  • Expression (planned for future) - controls vocal effort characteristics

  Voices and Voice Characteristics

  Installing new voices

  You can get other voices from the OGIresLPC download site. In particular, the AEC, TLL, and JPH voices should work well. There are others there for Mexican Spanish, British English, and German, but installing these is more complicated and requires some knowledge of Festival's internal workings. To install a voice...
  1. Download one of them, e.g., voice_aec_di-2.0.tar.gz.
  2. Unpack this in the SAME PLACE you unpacked the Flinger distribution.
  3. In a scm script like fl_test.scm, add a line
     (voice_aec_diphone)
     at the point at which you want the voice to change. To change it back to the default MWM, add a line
     (voice_mwm_diphone)
     

  Modifying vocal tract size

  Using a relatively simple signal processing trick, you can make the voice sound much deeper - this especially helps get a deep baritone singer quality. This works with any voice. Add the following lines to your scm script:
  (OGIresLPC.init++
(list '(vqual_mod
((vt_global_warp_wave 1.0)
(vt_voiced_warp_wave 0.90) ;;;<<<< change this
(vt_global_warp_lsf 1.0)
(vt_voiced_warp_lsf 1.0)))))
  You can change the "vt_voiced_warp_wave" line marked above to get various effects: a number between 0.0 and 1.0 will make the voice deeper, a number greater than 1.0 will simulate a smaller vocal tract (e.g., a child).

  Advanced Capabilities

  If you want to go further than what is given here, then the best strategy would be to learn about the capabilities of the Festival TTS system from the Festival manual. Flinger uses only some of the standard TTS modules, because singing presents a different set of language processing and signal processing issues than TTS does. The scheme code is editable by you -- see the files in festival\lib\, and in particular flinger.scm and ogi_effect.scm in that directory. In the future, we hope to allow a source code distribution so that others can customize Flinger on a deeper level.

  Licensing

  If you are interested in commercial applications of this technology, please contact Mike Macon at OGI.

  ---

  Please send constructive comments or suggestions to
  Mike Macon
  Last modified: Wed May 24 17:16:21 PDT