[eckythump]

The python speech server is on the downloads page. Extract it somewhere and run server.py from the speech.server folder. You can get command line options with --help, but you generally won't need them. This should work on any platform when set to have the server speak, and will work on windows and unix platforms for wav serving.
It uses the same backends as the addon, and you can select them from the addon.

---

Just FYI, I noticed today installed Cepstral voices are available in SAPI.

I took a look at the python server. Looks great. It's certainly a lot better constructed than my perl thing.

I saw that you were able to use the wave module to generate the RIFF/WAVE header. That's good. My header creation code seemed to work just fine, but I feared there might be edge cases that broke it. I trust the wave module to do it 100% correctly.

I have one suggestion for you. If you look at the code for my most recent httpttsd.pl, you'll see that I've made the generation of the AudioFormat value more dynamic, pulling from variables set right at the top of the file. You might want to do something similar so users can easily edit and set this value without needing to read through the code to find it.

I've had some experiences where some voices sound great at some frequencies, and awful at others. The infovox voices deserve special mention here, as they sound great at 22050, and bizarrely, sound awful at 44100, so it's preferable to be able to set that value easily should a user find that their voice of choice sounds like arse at our chosen defaults.

If you were feeling extra tricky, you could probably have "bits", "channels" and "samplerate" as optional POST variables so this can be set on the client side, falling back to a server default when not. I'm not sure I trust my makeshift WAVE header generation code to work with all the different permutations, but it's certainly an option you could consider.

Thankfully almost every voice available on Windows hooks into the SAPI system. Cepstral, AT&T Natural, Loquendo, RealSpeak, NeoSpeech, Infovox, etc. Even the windows port of eSpeak has SAPI hooks. The only voices I've seen that don't are Eloquence (comes with JAWS) and flite/festival compiled under cygwin (For all I know there're native Windows ports that do have SAPI hooks for these).

I played with Cepstral Callie on the pi. I noticed that swift has a -o option for writing out a wav. I tried it, but it was very, very slow. Too slow on the pi to be practical, but it might be an acceptable work around the ALSA requirement on other linuxes if it can generate the wav files fast enough.

I'll try and give your python server a whirl soon and see how it goes. I've got a standard install of Python 2.7 on my windows machine. Is this what you're using, or should I get some other more magical python package?