=== yofel_ is now known as yofel | ||
=== yofel_ is now known as yofel | ||
=== jalcine is now known as webjadmin_ | ||
=== webjadmin_ is now known as jalcine | ||
jalcine | Is HUD accessible over D-Bus? I'm working on a speech app that'd forward text as if it were typed into HUD | 19:00 |
---|---|---|
mhall119 | gord: ^^ | 19:05 |
mhall119 | tedg might know too, if you want to ask him when he's in tomorrow | 19:05 |
jalcine | mhall119: thanks :) | 19:19 |
jalcine | I might just pull in the code and just grep any references to d-bus and see where that takes me. | 19:19 |
AlanBell | jalcine: seen hud-cli and hud-dump-application and hud-list-applications and hud-verify-app-info? | 21:26 |
AlanBell | plus yes, the menus are very dbus exposed | 21:27 |
jalcine | I saw hud-cli, but not the others. | 21:27 |
AlanBell | I don't know what they all do exactly | 21:27 |
jalcine | And if d-bus is available, I'd stick to that. | 21:27 |
AlanBell | so what are you doing with it? | 21:27 |
jalcine | One less dependency to code in. | 21:27 |
AlanBell | I was thinking of using dbus to get the menu structure then mangling it into a jsgf grammar file to feed to pocket-sphynx | 21:28 |
jalcine | Well, we've gotten mediocre speech recognition going and right now, we're packaging the latest version of CMU's tools for acoustic model training. | 21:29 |
jalcine | Because the version in the repository aren't as proficient as the ones available. | 21:29 |
AlanBell | com.canonical.hud is the dbus thing | 21:29 |
jalcine | Yeah, SC's like powered on PocketSphinx, its it's battery. | 21:29 |
AlanBell | the dbus hooks for HUD are about submitting queries to it, should be controllable that way | 21:30 |
jalcine | That should be it, then. | 21:31 |
jalcine | I'd begin hacking up another plug-in. | 21:31 |
AlanBell | yeah, but it doesn't expose the full list of options through that | 21:31 |
AlanBell | you would need to poke at com.canonical.AppMenu.Registrar and call GetMenus() there, then go through the menus picking up entries, just like the HUD does | 21:32 |
jalcine | Well, all I really need to be able to provide input. | 21:32 |
jalcine | *able to do | 21:32 |
AlanBell | well how are you restricting the grammar? | 21:33 |
jalcine | We aren't. We're working on building speaker-independent models. HUD's a feature I'd like to connect with it. | 21:33 |
AlanBell | oh, well should be pretty easy then, but recognition accuracy won't be so hot | 21:34 |
AlanBell | if you can parse the menu structures and create a jsgf grammar file then pocketsphynx will only listen for useful words and will be more accurate in theory | 21:35 |
jalcine | It would, but we'd have to get the user to adapt that content into a specific model. | 21:36 |
jalcine | I proposed a delicately difficult idea: http://thesii.org/wiki/SpeechControl/Ideologies/AutomaticAcousticModelImprovement | 21:38 |
nava | Hi ,what do you think about this feature for unity ? http://brainstorm.ubuntu.com/idea/29302/ | 21:38 |
jalcine | But I figure with HUD + SpeechControl, AAMI (above) would gradually improve in use. | 21:38 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!