http://hts-engine.sourceforge.net/
This version of flite_hts_engine has been compiled on DebianDog_jwm wheezy; dotpet packaged using Puppy Slacko 6beta. I have patched it to output to stdout so that it can drive, for example, aplay via a pipe. I am still working on the code, but will upload a stable version of the patches once my testing is complete.
As provided here, this program, buffers and plays a maximum of 1024 characters, but the program could be compiled to use a larger buffer, albeit with a larger processing start delay. That is a disadvantage compared to standard flite, which has virtually no startup delay or buffer limitation. However, standard flite has much poorer voice quality. The previous version of flite_hts_engine, which I compiled and released in 2009, tended to garble certain combinations of text, but that problem happily seems to be resolved in this later version, though more testing is required to confirm that.
The provided package include flitet, flitetf, flitet2wav, and flitetf2wav. These are very small exemplar shell scripts which the dotpet installs in /usr/local/bin. They are intended as simple exemplars only, as an aid to developers wishing to create their own more sophisticated driving scripts. However, these scripts are perfectly usable as is. Open them in any text editor to see brief, simple, usage instructions, which are repeated in briefer form here (or enter, for example, flitet --help):
Usage:
Example of use: flitet "hello world" "See you later" | aplay
Example of use: flitetf infile1.txt infile2.txt | aplay
Example of use: flitet2wav "hello world" "output.wav"
Example of use: flitetf2wav "infile1.txt" "outfile.wav"
flite_hts_engine is a special version of flite which has been patched for use with HTS voices. The HTS voice provided in the attached package is trained using Hidden Markov Model from the CMU ARCTIC database. Festival TTS could also use CMU ARCTIC, but that required a 100 MB download of the voice data... However this technique of using HMM results in pretty much as good quality with less than 2 MB of voice data (uncompressed)! Note that there quite a number of alternative HTS voice data sets now available via links from HTS website and by googling. I haven't tried these as yet, however, and don't know how easy or possible it will be to get them to function with the supplied scripts. Some relates HTS projects look interesting, however. For example:
http://sinsy.sourceforge.net/
http://sp-tk.sourceforge.net/
http://homepages.inf.ed.ac.uk/jyamagis/ ... age54.html
The following may be particularly interesting:
http://mage.numediart.org/
For additional parameters that can optionally be supplied to flite_hts_engine
(e.g. -fm 4 -a 0.6), enter on commandline:
Code: Select all
flite_hts_engine -help
flite_hts_engine in, for example, supplied script flitet.
For a while at least you can download the packages (deb or dotpet) from the dropbox links provided below:
For DebianDog (tested on jwm version):
https://www.dropbox.com/s/jyjx9g3j4jssc ... 6.deb?dl=0
For Puppy (tested on Slacko 6beta):
https://www.dropbox.com/s/12j5dil65m3z9 ... t.pet?dl=0
William