Festival does in fact run under windows. It's something of a pain
to compile, but there are also binaries available. As for performance
vis-a-vis L&H, you'll have to decide which you prefer :) Festival
uses concatenative synthesis and is a full platform; you can modify
everything, add new pronunciations, etc. The L&H synth in MS Agent
is a formant-based synth (classic buzzy sound). Some people prefer it,
others prefer the artifacts of concatenative synthesis. In any event,
Festival is much more flexible than the L&H system because it's
open source/free software.
See also http://www.festvox.org for information on how to build new
voices for Festival, something you'll never get from L&H.
I can post the binary location for Festival on Windows if there's
interest...
kevin
On Wed, Mar 28, 2001 at 02:34:49PM -0700, Wilkinson, Mike wrote:
>
> ...And I was just building an MS agent extention to interface with POE...
> Any chance this thing will work under windows using the a festival build for
> windows? Any idea how it performs compared to the L&H engine ms agent
> provides?
>
> - MW
>
> -----Original Message-----
> From: David Huggins-Daines [mailto:[EMAIL PROTECTED]]
> Sent: Wednesday, March 28, 2001 8:45 AM
> To: [EMAIL PROTECTED]
> Subject: [ANNOUNCE] Speech and audio POE components, modules
>
>
> Hi,
>
> For the last few months, we here at Cepstral have been working on
> building speech input and output applications using Festival, Sphinx,
> Perl, and POE (of course). Now, it seems, the modules underlying them
> are finally ready to be released into the unsuspecting world. Behold
> our works, ye mighty, and despair:
>
> POE::Component::Festival - POE component for speech synthesis
> -------------------------------------------------------------
>
> This is a servlet component which allows you to communicate with a
> Festival server in event-driven fashion. It is essentially an adaptor
> for the Festival::Client::Async module, which can also be found at the
> site below. You must have Festival
> (http://www.speech.cs.cmu.edu/festival/) installed and running as a
> server to use it.
>
> POE::Component::SPX - POE component for speech recognition
> ----------------------------------------------------------
>
> This is a servlet component which interfaces to the Sphinx-II speech
> recognition system. It is an adaptor for Speech::Recognizer::SPX,
> which, again, can also be found on our site. Currently, you must have
> a recent version of Sphinx-II (http://www.speech.cs.cmu.edu/sphinx/)
> from CVS (see http://www.sourceforge.net/projects/cmusphinx/) - the
> 0.2 release is too old.
>
> POE::Component::Audio - POE component for audio input/output
> ------------------------------------------------------------
>
> This is a servlet component which provides an asynchronous,
> event-driven, quasi-real-time interface to OSS audio devices (such as
> on Linux or *BSD). It requires our Audio::OSS module.
>
> POE::Component::SilenceFilter - POE component for audio silence filtering
> -------------------------------------------------------------------------
>
> This is a servlet component which is interposed between
> PoCo::Audio::Input and PoCo::SPX in order to remove silence regions
> from audio input and trigger the starting and stopping of utterance
> processing. It adapts Audio::SPX::Continuous, which is part of the
> Speech::Recognizer::SPX distribution.
>
> All of these modules are released under the same license as Perl
> itself, and the underlying software (Sphinx-II and Festival) is
> released under BSD-like licenses. So yes... it's all free software!
>
> You can obtain all these modules, and their dependencies, from our
> developers' site at
> http://www.cepstral.com/website/pages/cep_developers.html
>
> Some caveats:
>
> Currently this stuff has only been tested under GNU/Linux on the i386
> platform. Due to the dependence on OSS audio (or OSS emulation, such
> as in ALSA), the audio code is only likely to work on Linux and *BSD
> systems with OSS drivers. The synthesis and recognition functions,
> however, should be sufficiently platform-independent.
>
> The modules and components appear to be quite stable, though there are
> many features remaining to be implemented.
>
> All modules and components are documented, and all except
> PoCo::SilenceFilter have test suites. However, test suite and
> documentation coverage is not complete, and there should be more
> example code.
>
> Please send suggestions, bug reports, success and failure stories, and
> questions to me, <[EMAIL PROTECTED]>.
>
> A bit more about us:
>
> Cepstral LLC was founded in 2000 by Kevin A. Lenzo and Alan W Black,
> leading researchers in speech recognition and synthesis, to develop,
> market, and support open-source speech technology. We specialize in
> high-quality characteristic speech output systems, based on proven,
> scalable open-source engines.
>
> --
> David Huggins-Daines | [EMAIL PROTECTED]
> Toolsmith | http://www.cepstral.com/
> Cepstral LLC | We Build Voices