New audio user interface

psypete · 08-19-2005, 06:39 PM

I was way too cheap to buy an LCD to control my CarPC, so I wrote this to bridge the gap. Basically if you have mplayer and flite or festival (flite works better) and a keyboard you can control the PC using only a keypad and no screen. As you pick the plugin to use and move the keys the speech synthesis software will speak to you where you're going and what you can do.

Of course it's plugin-based and there's only 1 plugin at the moment (for playing MP3's) but if you just need something quick and easy this will work. Just set up Linux to make the first terminal the path to this program and configure the path to the audio or video files.

The end result is you can get a cheap USB or PS2 number pad and mount that somewhere in your car and browse your MP3's without ever taking your eyes off the road. This will also work in laptops for example, so no more dangerous side-glances at the laptop to pick an MP3 or change a window.

If you find it useful but want plugins made to control other things just let me know and i'll see if it's not too difficult to do. With some additional script I have you can auto-mount CDROMs and SCSI mass-storage devices and access it all through CAUI. Maybe I should write up a small tutorial...

pkg · 08-21-2005, 04:10 AM

I would love to hear about how you did this. I think voice synth can go a long way in other interfaces too.

psypete · 08-22-2005, 10:51 PM

Quote: Originally Posted by pkg

I would love to hear about how you did this. I think voice synth can go a long way in other interfaces too.

Well the theorey is pretty simple... have the computer read back to you whatever is supposed to be on the screen. There's already bunches of software to just read everything on the screen but I needed to only read back the file names and directories.

The actual text to speech is easy with festival or flite. Just pass them words on the command-line and they speak 'em. You can usually also pipe text to them. So all you need to do is capture keystrokes on a keypad, figure out what to do (read the next file, previous file, go into a directory, make a playlist, etc) and when it's time to speak call the TTS program with the words to speak. This capturing of keys and deciding what to do is what CAUI does.

Of course you'll also need a music player like MPlayer and some extra pieces of software probably to make it all come together easy. So install flite, mplayer, caui, and the Term::ReadKey perl modules in your Linux machine and you're mostly there. Then just configure your distribution (in /etc/inittab) to run CAUI as the first terminal when your system starts up; with a USB keyboard or numpad plugged in you should be able to start using CAUI right when the PC boots up.

You'll also need to edit /etc/caui.conf for the location of the music but thats about it as far as I remember. Any other questions?

P.S. Would anyone like a copy of my CarPC linux distribution? It's ~110MB so it fits nicely on a CompactFlash card and can be run read-only so you don't have to shut down your PC safely.

TheLlama · 09-26-2005, 01:35 PM

Good idea. Now I'm seriously thinking about using Festival in my player also.

intuitionsys · 09-26-2005, 02:01 PM

Although I'm using a display as well I've been using festival for quite some time with zero problems. I'd like a larger choice of voices but it works fine. Maybe I'll check out flite too...

So far the handiest thing I've found for voice response (besides the obvious coolness factor

) is moving through the playlists as I currently have 100. Moving through that many while driving could be dangerous.

TheLlama · 09-26-2005, 02:03 PM

intuit- so how do you have it set up? If a list element is selected (highlighted) for X amount of time then it will announce the selection? Or does it just start speaking each time you select an item? Just trying to invision what your system is actually doing.

intuitionsys · 09-26-2005, 03:02 PM

Basically I press the playlist button on the remote, it displays the next playlist if it exists, and immediately starts a 1s timer, but only if that timer instance isn't already running, effectively becoming it's own boolean indicator for whether or not to speak the current playlist:

Snippets from the Silverwolf code:

Code:

...

    qtPLDelay = new QTimer( this );
    connect( qtPLDelay, SIGNAL( timeout() ), this, SLOT( slotSayPlaylist() ) );

...

void MP3LibDlg::slotPlaylist()
{
    PlaylistDisp();
    if( qtPLDelay == 0 )
        return;
    if( qtPLDelay->isActive() )
        qtPLDelay->stop();
    qtPLDelay->start( 1000, TRUE );
}

...

void MP3LibDlg::slotSayPlaylist()
{
    QString qsSay;

    Speak( qsSay.sprintf( "Playlist %d", iCurrPlaylist ) );
}

Of course I left alot of stuff out but I hope that makes sense.

TheLlama · 09-26-2005, 03:38 PM

Yep, thats pretty much how I figured you would do it.

mofu · 09-27-2005, 11:24 AM

What do you think the the chances of getting a Navigation program like RoadNav working though a voice output interface.

Lets assume you have some address enter capability via a 4x20 LCD and keyboard (mini). Of course, a GPS receiver on serial or USB.

The program should output simple text instructions to the LCD and voice module. "Turn right in half a mile . . . "

LCD Touchscreens look nice when your parked, but have serious functionality issues once you are actually driving. So I think you approach has a lot of merit.

TheLlama · 09-27-2005, 01:13 PM

It wouldn't be hard to do if you already have the source to do the rest of the work. I'm considering doing something like this _eventually_. I agree about the touchscreens looking poor while in a moving car, especially when the sun is low. Thats why VFDs are bangin'.