Check this out -
http://www.mp3newswire.net/stories/2001/mxp100.html

It looked like a review of a boring, run of the mill portable mp3 player at first, but then they said the magic words - "voice recognition" - and they had me excited. I don't know about the rest of you, but for me the deciding reason to purchase an Empeg was because of its ability to search for songs directly from the remote. Now, this little player sounds like it's supposed to be able to jump straight to a song if you simply say it. If so, sign me up.

-Adam

Now, what I've been wondering is how long before we could implement this for the Empeg? How cool would it be to say something like "Play Dave Matthews Band" and it creates a playlist of DMB for you.

Am I invisible?

This does require a VR engine. This handheld player uses the Lucent engine, which is one of the ones we've been looking at.

Obviously, implementation would require licencing...

Hugo

Those guys aren't the only ones. I'm aware of another mp3, hard drive based, in car player that will have voice recognition. Very good voice recognition, from the sounds of it, like, back seat windows open voice recognition.

Calvin

...ah... [smiles] ...the Lucent engine...

(...what our studio audience does not know is that our guest works for a major American Telephone and Telegraph company, and has seen this engine perform its stuff during many proprietary internal demonstrations...)

I would really like to see a VR controlled empeg!!! I know that nothing has been made available, but are you willing to share your experiences (i.e., have you ever tried to prototype VR with the player software, and how did it work)?

Hugo,

How much would a VR license cost? $1K, $10K, $100K?
I'm sure that it matters how many instances.
Is it something the Sonic could use for other projects as well?
If the cost was nonminal ($50) per user maybe some of the empeg/rio owners could pitch-in?

Have Fun,
BillA.

I agree that VR is something that would be worth the extra cash. I would be willing to pay for it. ($50)

Sean

I agree that VR is something that would be worth the extra cash. I would be willing to pay for it. ($50)

I'd be willing to pay more than that.

tanstaafl.

>>I agree that VR is something that would be worth the extra
>>cash. I would be willing to pay for it. ($50)
>I'd be willing to pay more than that.

What would the video recording of you, sitting on your Bikempeg - seemingly talking to yourself - cost us them? :-)

We are assessing a few different systems. At the moment this work has to fit around scheduled projects but with luck we may get an OEM interested in the functionality (which will provide more momentum).

There are several impressive VR systems on the market now. The critical issues are triggering (voice triggering is flakey at best) and size of vocabulary (a tag per artist stretches most of the systems to the limit with a large music collection - tags for song titles are out of the question).

Rob

How about spell-by, augmented by visual interface cueing? Kind of like the search screens, you say a letter, it hones down the choices, and displays them.

Calvin

What would the video recording of you, sitting on your Bikempeg - seemingly talking to yourself - cost us them? :-)

Nothing unusal about that -- I talk to myself all the time. Seems like I'm the only one who ever listens...

tanstaafl.

There's nothing abnormal about talking to yourself.

The line is crossed when you reply to yourself with "huh?"

I think that is very contrary to the basic intentions of voice recognition. You should never need to look away from the road - perhaps your product doesn't even have a screen. The ideal solution is for you to talk to your stereo and for it to talk back to you.

Rob

Okay how about this... Speech recognition is hard, but text-to-speech is easy. So, along the lines of the "spell by" thing... The operator spells the (artist | title | source | year | genre) they want to input and when the player has an exact match it speaks the exact match back to the user. No need to look at the interface.

I'm brainstorming non-ideal options though because you called out a non-ideal recognition engine. Other possibilities are: Playlists only, but some people have enormous numbers of playlists. Pin numbers only. Controls only (louder, software, power on, power off), no title search. Fuzzy title search --> "Sounds Like" this? E.g., you don't have to have the player match a specific title to a specific song. You utter your intentions, and it goes and assembles a playlist of songs that resemble what you said.

Yep, I'd rather have it ideal though, agreed.

Calvin

Voice synthesis is often not done "right" -- often it ends up being infuriating.

Calvin

yn0t_ -

Or, along those same lines, have it give an audio cue when the list is down to something managable, ~5 entries, then you can have it read them off to you one by one (with something like "next" and "name" commands) or you can keep on spelling, and then like you said when it gets down to one it would read it off to you.

Also, I agree wtih eteralsun - this isn't my ideal of a voice recognition system, but on the plus side it seems (with my limited knowledge) plausible for what's available.

-Adam

Eh, it's gotten real good lately. Not perfect but I don't care if it can't exactly pronounce a word right, I just want to know when it's found an exact match, and what that match is...

I'm not saying that voice synthesis itself is terrible or not up to snuff. It's the application of synthesized voices that infuriates. I think the forefront of voice synthesis applications are in the interactive voice response systems (IVRS) and everyone knows how bad those are! Even consider the ones with integrated speech recognition. ("Say or press 1..."). E.g., ("Your portfolio contains the following stocks.....").

Calvin

We've looked quite closely at TTS (text to speech) systems recently, and on the whole embedded systems are still not particularly good. Large memory hogging systems, on the other hand, are starting to sound extremely impressive.

Rob

Absolutely. I haven't heard anything specific about the TTS-on-a-Chip, but server-based TTS is getting more human-sounding all the time. If you've ever been to AnaNova, you'd hear what I mean. We use the same engine as that site (L&H Realspeak) in some of our company's IVRs, and they do sound pretty realistic.