dosnet



Setting up Festival and using better voices
--------------------------------------------------------------------
Section 1: Get Festival up and running with basic voices

1) Additional reading:
Here are some places I pulled notes from:
http://gentoo-wiki.com/HOWTO_speechd (speechd and festival notes - mbrola too)
http://forums.gentoo.org/viewtopic-t-195579-highlight-festival.html
http://dailypackage.fedorabook.com/index.php?/archives/42-Productive-Monday-Festival-Speech-synthesis.html

#Someone elses notes on how to build your own festival voice based on recordings of your voice:
http://www-csli.stanford.edu/semlab-hold/muri/system/festvox.html


2) You definately need to use Festival 1.96 or better, the older version sound very poor:
http://festvox.org/packed/festival/latest/

#Get these two packages to start:
festival-1.96-beta.tar.gz
speech_tools-1.2.96-beta.tar.gz

Unpack these tars in the same parent directory, festival will unpack into a directory called "festival", speech tools into "speech_tools". Compile speech_tools first, then compile festival. Next unpack these other packages in the same parent directory (these get loaded into directory "festival").

festlex_CMU.tar.gz
festlex_OALD.tar.gz
festlex_POSLEX.tar.gz

3) Next are the voice packages:
festvox_cmu_us_awb_arctic_hts.tar.gz
etc.

4) These packages help the voices to sound MUCH better:
festvox_kallpc16k.tar.gz
festvox_kedlpc8k.tar.gz
festvox_kedlpc16k.tar.gz


5) Now edit this file to use the new voices:
./festival/lib/siteinit.scm

;And add this line:
(voice_kal_diphone)
;And then change the line like this to your new voice (notice the prepended "voice_" to the voice name):
(set! voice_default 'voice_nitech_us_clb_arctic_hts)


This should have festival working now with voices that can be changed.

6) Voices:
; pansy male voice      = cmu_us_bdl_arctic_hts
; deeper male voice     = cmu_us_jmk_arctic_hts
; female voice          = cmu_us_slt_arctic_hts

; Much better voices (from Japanese site, see section 2:
; female voice          = nitech_us_slt_arctic_hts
; another female        = nitech_us_clb_arctic_hts
; light male voice      = nitech_us_rms_arctic_hts
; deeper male voice     = nitech_us_jmk_arctic_hts


echo "Mary had a little lamb." | festival/bin/text2wave -o test1.wav



--------------------------------------------------------------------
Section 2: Get the very good English voices from Japan

The HMM-based Speech Synthesis System (HTS) has been developed by the HTS working group of Japan and makes Festival understandable enough for serious use (comparable to the 162MHz weather radio text-to-speech voice).

# More information:
http://forums.gentoo.org/viewtopic-t-195579-highlight-festival.html

Look for "HTS voices", download and unpack them:
http://hts.sp.nitech.ac.jp/?Download#k1dc0e56

# Nitech US AWB ARCTIC HTS
# Nitech US BDL ARCTIC HTS
# Nitech US CLB ARCTIC HTS
# Nitech US JMK ARCTIC HTS
# Nitech US RMS ARCTIC HTS <-- used by our radio station
# Nitech US SLT ARCTIC HTS


These packages unpack into a new directory called "lib", do this to copy them into the correct festival location (assuming you are in the parent directory with "festival" and "speech_tools":
cp -R ./lib/* ./festival/lib/

Now edit the Festival config file siteinit.scm and try them out.