Fixes to make the project work with OpenFST 1.8.3 #89
Conversation
Merge pull request AdolfVonKleist#65 from danijel3/master
|
I do have a fork where I merged some of the fixes made in various PRs here. I used it to publish a PyPI package with the prebuilt Python bindings and all necessary OpenFST binaries. I could update that to OpenFST 1.8.3 |
|
The reason I did this is because I used Phonetisaurus extensively with Kaldi and they moved on to 1.8.3. I do recommend the update for standards adherence, but it feels to me that there are probably more changes worthwhile than the couple I made here. Oth, I've been using it in this state for a few hours and so far so good... |
|
Thanks for the fork. If you need help with anything just let me know. |
|
Have some update this? Phonetisauru worked? I go develop a TTS Engine for Android using Flite, I go switch g2p because trained of g2p (Random Forest) of Flite is very slower. Because is trained using 1 core of CPU. |
CXXFLAGS "-ffast-math" in phonetisaurus-g2pfst not generate pronunciation for all word
Update utfcpp library and fix error of nil pronunciations for all words
Add support for use KenLM for train ngram model
Fix args for train ngram model using kenLM.
Support for KenLM
3 participants
Most of the world has moved to version 1.8.3 of OpenFST. I made some cosmetic changes to allow the project to compile correctly.
Note, there are still a ton of warnings during compilation. I haven't made thorough checks to see if everything works (it seems to), but it does compile everything.