OpenMRCP (2)
I finally managed to have a look at OpenMRCP.
Installation
Compilation and installation goes smoothly enough - as long as you have the right versions of everything, in the right places (e.g., sofia-sip has to be in /usr/local/include/).
Fake TTS/ASR demos
The OpenMRCP webpage shows how to use OpenMRCP client and server for TTS and ASR. No real synthesis or recognition is done - the TTS demo plays an audio file you give it, and in the ASR demo the client sends the server an audio file and the server sends back the message ‘Recognized’ - but the point of the demo is to show the SIP/SDP/RTP/MRCP infrastructure in action, which it does very nicely, printing out the control session messages sent and received.
To have real TTS or ASR you need to provide the resource plugins yourself. Cepstral’s Swift TTS engine can be used for the speechsynth synthesis resource, but there seems to be very little to show how to write a speechrecog resource.
Forum
There is a forum on the OpenMRCP website, but this is very quiet, partly because registered users are not allowed to start new topics. [UPDATE: This now (051107) seems to be fixed.]
Aim
Given that the infrastructure is so well looked after, it seems that a sensible medium term aim for me would be to write a speechrecog resource plugin which would enable speechrecog resources based on HTK or Sphinx3 acoustic models. This would be a medium rather than a short term aim because:
- I have a few other things on at the moment;
- the plugin would be written in C, interfacing between two other programs (OpenMRCP, and HTK or Sphinx3) also written in C;
- my C is very basic.
Todo
My first steps will be:
- learn more about C;
- find or write explicit specifications of the interfaces of OpenMRCP, HTK and Sphinx3. With the latter two, especially about accepting media streams.