TTS API Provider and Retrieving synthesized audio data

[Hynek Hanke]

Hello all,

There is currently no support for audio retrieval in Speech Dispatcher,
although there might be ways around in specific cases (as suggested
in other posts).

Audio retrieval is partly implemented in TTS API Provider.  Since it
has much cleaner design, the only difference in the two cases really
is whether the output module sends audio to the TTS API Provider Audio 
Interface
or to a custom destination specified by the client. The audio data are being
sent asynchronously over a newly created channel, so one of the
benefits is that the receiver (Audio Interface or client application)
starts receiving data as soon as any data is synthesized, not only
after the whole synthesis is completed.

It is also possible, if the synthesizer implemented TTS API directly
(which is already partly true in case of eSpeak and should be well possible
for Festival or other open source tools), to send audio data directly
to destination (e.g. Pulse or client), not through the output module. This
should also significantly speed up the whole process.

TTS API Provider is not currently finished (not in the installable-usable
phase yet), but has gone quite far in many aspects. Experiments have shown
that it is not necessarily slower than the C-based Speech Dispatcher, 
because
while python is a bit slower in code processing, Python allowed to easily
implement much better design. It's a pitty it has come to standstill.

For the future of C-based Speech Dispatcher, I think TTS API (the 
specification,
not the python implementation) is a very good inspiration, because we 
thought
thoroughly and discussed many of these issues already.

See the TTS Audio Retrieval protocol spec:

    http://cvs.freebsoft.org/doc/tts-api/tts-api.html#Audio-Retrieval

(the essence of it, the specific implementation over text TCP protocol
is just an example)

Anyway, please keep up the very good work!

With regards,
Hynek Hanke
Brailcom, o.p.s.