[Freeswitch-users] limits on text to speech output length
daniel.neubert at solomo.de
Thu Sep 23 21:41:27 PDT 2010
It's not only good for having a better scalable infrastructure - it
also gives you the ability to freely change your tts engine or add
functionality like speech recogniction or recording.
Am 24.09.2010 05:59, schrieb A. Lester Buck III:
> Dear Daniel,
> Thanks so much for the pointer. I'll certainly check that out. I don't
> really have much of a scaling issue at the moment, but I sure hope I do
> someday! Always good to keep the architecture as small parts loosely
> Best regards,
>> instead of calling swift directly you can consider calling it via an
>> MRCP server like UniMRCP.
>> It's all free software and works like a charm in my setup. I'm using
>> mod_unimrcp on our FreeSWITCH servers and am running UniMRCP as a
>> server on a single node that is our dedicated speech server.
>> That scenario also offloads the task of speech generation to another
>> machine to save resources on the FreeSWITCH nodes. If you use other
>> cepstral languages than english you need to manually patch the sources.
>> Am 23.09.2010 22:57, schrieb A. Lester Buck III:
>>> I'm running Freeswitch built from trunk a week or so ago, 1.0.head.
>>> I am using a Cepstral voice for text to speech. I generate a long
>>> string of text, including SSML breaks, that is rendered to speech by the
>>> swift utility packaged with Cepstral voices. The test string I built
>>> renders as 1m:46s of speech, and a .wav file about 1.6MB, from the swift
>>> command line.
>>> I changed single quotes to double quotes and escaped them, as mentioned
>>> on the wiki page for mod_cepstral, and I now have a working text to
>>> speech path using ESL built for Ruby 1.9.2, using the sample Ruby code
>>> on the wiki.
>>> Everything basically works fine, even on my test machine which is a
>>> $5/mo Xen VPS with 10ms kernel timer resolution. (lowendbox.com)
>>> My issue is that the speech cuts off midway through the call, when I
>>> guess some limit on generated output is being reached. Before I dig
>>> into the guts of mod_cepstral and Freeswitch, could someone point me to
>>> what might be limiting the length of output? Swift seems to generate a
>>> wav file of indefinite length, so at least the issue is somewhere in the
>>> open source parts of the server.
>>> Having yet looked at the guts of mod_cepstral, I presume that a
>>> high-performance output module does not have the swift utility write a
>>> .wav file to disk and then play it back, but instead lets swift shove
>>> the bytes directly into a socket. What is a good Freeswitch example of
>>> this architecture?
>>> FreeSWITCH-users mailing list
>>> FreeSWITCH-users at lists.freeswitch.org
>> FreeSWITCH-users mailing list
>> FreeSWITCH-users at lists.freeswitch.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the FreeSWITCH-users