Thanks Marc, I agree with you that quick start/stop times are needed.  However, I am still trying to understand what you mean by latency.  I thought quick start time equals low latency.  Can you describe a scenario with a quick start time but (relatively) long latency?  Thanks, Pete ===== Pete Brunet, (512) 838-4594, TL 678-4594, brunet@us.ibm.com, ws4g IBM Accessibility Architecture and Development, 11501 Burnet Road, MS 9026D020, Austin, TX  78758 <table width=100%> <tr valign=top> <td width=40%>"Marc Mulcahy" <marc@plbb.net> 03/07/2005 02:11 PM <td width=59%> <table width=100%> <tr> <td> <div align=right>To</div> <td valign=top>Pete Brunet/Austin/IBM@IBMUS, <accessibility@lists.freedesktop.org> <tr> <td> <div align=right>cc</div> <td valign=top> <tr> <td> <div align=right>Subject</div> <td valign=top>RE: [Accessibility] RE: Multimedia framework requirements</table> <table> <tr valign=top> <td> <td></table> </table> pete, it's just another case of needing the ability to start and stop speech immediately-- not a case of a need for low-latency audio.  the scenario is:   * Speech is talking * User presses a key * Speech is interrupted (could be by halting DMA and/or resetting the sound card) * key is echoed (synthesis is started and audio starts streaming to the audio device)   note that hardware latency isn't a factor in either the start or stop scenario-- in the stop scenario, it's more a factor of how fast the sound card can be reset (DMA halted).  this has nothing to do with hardware latency.   in the start case-- the perceived response time for the user is based on how fast the sound card can start transfering data from RAM to the hardware audio buffer, not a factor of how big the chunks being transferred are.   When you start mixing in software, the latency of the software mixer does play a factor-- since the stream to the soundcard is then continuous.  but when characterizing the accessibility requirement, I think specifying low-latency is the wrong terminology-- what we need is quick start and shut up times.   Marc -----Original Message----- From: accessibility-bounces@lists.freedesktop.org [mailto:accessibility-bounces@lists.freedesktop.org]On Behalf Of Pete Brunet Sent: Monday, March 07, 2005 12:24 AM To: accessibility@lists.freedesktop.org Subject: [Accessibility] RE: Multimedia framework requirements Marc, How does the need for instant echoing of keyed characters when entering text fit in with this situation?  Thanks, Pete ===== Pete Brunet, (512) 838-4594, TL 678-4594, brunet@us.ibm.com, ws4g IBM Accessibility Architecture and Development, 11501 Burnet Road, MS 9026D020, Austin, TX  78758 ---------------------------------------------------------------------- Date: Sat, 5 Mar 2005 17:55:27 -0700 From: "Marc Mulcahy" <marc@plbb.net> Subject: RE: [Accessibility] Multimedia framework requirements To: "Gary Cramblitt" <garycramblitt@comcast.net>,                 <accessibility@lists.freedesktop.org> Message-ID: <KKEGJCDELINGIGICHANAGEKBEDAA.marc@plbb.net> Content-Type: text/plain;                 charset="iso-8859-2" Well, for what it's worth, here is my $.02. 1. We in the accessibility community will never succeed in trying to re-invent the multimedia server.  There have been many attempts by people with expertise in multimedia with varying degrees of success.  So I think the right approach is to focus on selecting an existing solution that comes closest to what we need, and either living with it, or proposing changes which will bring it closer to what we need. 2. The biggest oversight in gnome-speech was that it did not directly handle the audio coming out of software synthesizers.  Given my experience with several commercial and open source speech engines, I came to the conclusion that the speech framework *must* have control over the audio samples and where they go.  If we leave it up to the speech engines, they will all implement things differently, and we have much less of a good chance of providing a good experience for the end user.  Having control over the audio gives us better control over quick startup and stop times, as well as the ability to route speech to diferent destinations-- files, headsets, speakers, telephone lines, etc. 3. To my mind, ALSA comes the closest to what we need in an audio framework on Linux.  It's now standard, and provides methods for mixing audio streams on soundcards which can't do it in hardware.  The prioritization of audio-- I.E., muting the MP3 player when the computer needs to speak something or when a user receives an internet phone call, is the only piece which appears to be missing. Another audio-related aside...  I think there's been some mischaracterization of a requirement.  Everyone seems to suggest that what we need is low-latency in an audio server or environment, and I'm not convinced that this is the case.  You need low-latency, or at least good synchronization, if for example you want to animate a character using text-to-speech as the voice.  But, I think from an accessibility point of view, what we really need is quick start and shut up times, not necessarily low latency, although low latency is better.  For example, from a blind usability point of view, I don't care if the ap sends the sound card a 128 KB buffer of audio or a 1 KB buffer of audio, as long as the sound stops immediately when I press a key, and as long as it starts immediately when there's something to be spoken. My experience shows that low-latency is in fact not necessarily desirable when working with speech.  Presumably speech is a background process which goes on while other more intensive tasks are happening in the foreground-- copying a file, filtering audio, or something of that sort.  the lower the latency, the harder it is to keep speech happy in the background, especially during periods of high disk activity or network load. Rather than having to feed the soundcard 1 K blocks of data, I'd rather synthesize 64 K of data, and dump it to the sound card, and let the DMA controller transfer it while the processor does something else.  and as long as I can shut it up immediately, the user doesn't know the difference. Marc -----Original Message----- From: accessibility-bounces@lists.freedesktop.org [mailto:accessibility-bounces@lists.freedesktop.org]On Behalf Of Gary Cramblitt Sent: Saturday, March 05, 2005 6:20 AM To: accessibility@lists.freedesktop.org Subject: Re: [Accessibility] Multimedia framework requirements On Friday 04 March 2005 03:43 pm, Hynek Hanke wrote: > 2 Audio requirements You may want to think about supported audio formats.  Most existing synths seem to produce .wav files (Microsoft riff) or .au. Also, there's the issue of how to deliver the audio to the audio framework. Streams could be more efficient than files?  The TTS API discussion has this as an unresolved item. -- Gary Cramblitt (aka PhantomsDad) KDE Text-to-Speech Maintainer http://accessibility.kde.org/developer/kttsd/index.php ------------------------------