- From: Al Gilman <Alfred.S.Gilman@IEEE.org>
- Date: Thu, 27 Apr 2006 14:06:17 -0400
- To: "Shane Smith" <safarishane@gmail.com>, www-voice@w3.org
At 11:37 AM -0500 4/27/06, Shane Smith wrote: >Hey Folks, > >One of the best features of audio is the ability to play backup tts >should the audio source be unavailable. Currently though, <audio/> >requires either the src or expr attributes to be listed, or a badfetch >is tossed. Well, I've come upon a scenario where I wouldn't >necessarily want to list either, and use the backup tts feature in a >way that wasn't anticipated, but could be very useful. > >For example, let's say I'm playing an account number to the caller: > <audio expr="AudioDirectory+'1.wav'">1</audio> > <audio expr="AudioDirectory+'2.wav'">2</audio> > <audio expr="AudioDirectory+'3.wav'">3</audio> > <audio expr="AudioDirectory+'4.wav'">4</audio> > <audio expr="AudioDirectory+'5.wav'">5</audio> > <audio expr="AudioDirectory+'6.wav'">6</audio> > >I'm not going to prerecord every possible number, so I play audio one >digit at a time. But, if for any reason one or more of them is >unavailable, I would rather the whole thing be read back as TTS. The >above code sounds horrible as backup tts, and I do not really have >ssml control over it. > >What I would like to see is this: > ><audio> > <audio expr="AudioDirectory+'1.wav'/> > <audio expr="AudioDirectory+'2.wav'/> > <audio expr="AudioDirectory+'3.wav'/> > <audio expr="AudioDirectory+'4.wav'/> > <audio expr="AudioDirectory+'5.wav'/> > <audio expr="AudioDirectory+'6.wav'/> > <prosody rate="-10%"> > <say-as interpret-as="digits">123456</say-as> > </prosody> ></audio> > >This (or something similar) would allow you to chain a bunch of audio >prompts together, but if any of them fail, have a single backup tts >prompt replaced for all of them. In my head, if all numbers wav files >were unavailable, or even if just 5.wav were unavailable, none of them >would play, and the ssml'ed TTS would play instead. > >Thoughts? Option 1: have you looked at using a SMIL[1] as the @src value? I think you would have to do things with test variables to force it to fail atomically if any of the embedded <seq> constituents were unavailable, but the concatenation is there. Option 2: Alternatively, review DISelect when it re-emerges in its next incarnation [2] and consider how you would fold this into VoiceXML 3.0. Option 3: Program the sequences-and-options logic in SCXML [3]. You may get this in VoiceXML 3.0 even if you don't push for it. Al [1] http://www.w3.org/TR/2005/REC-SMIL2-20051213/ [2] (some version dated after today of) http://www.w3.org/TR/cselection/ [3] http://www.w3.org/TR/2006/WD-scxml-20060124/ Al > >Thanks, >Shane Smith
Received on Thursday, 27 April 2006 18:06:32 UTC