First, one more clarification, since you seem to have my mixed up with the guy thinking of selling the $600 unit. My program, Voodoo Voice, is software only and works as a plugin for Winamp. Its available now, at the URL in my info line above, and is free (beer), although I would say its still beta at this point.
Unfortunately, it does require you to say the entire, exact name of the song. I've been trying to come up with a way around this, but haven't had much success. For now the best solution is to give your songs easier names.
Voodoo Voice works through Microsoft's Speech API which simply defines an interface between voice applications and engines. I've done most of my testing with Microsoft's freely available engine, but if you have NaturallySpeaking or ViaVoice it will use that engine instead and you will probably get better results.
Check out the URL above. Voodoo Voice does exactly what you're saying. Its software only, and runs in Windows at that, but if the original guy in this article has complete hardware capable of speech recognition for $600, perhaps he/you could simply install Windows and use this voice plug-in instead.
(Yes, shameless plug, since I'm the author of Voodoo Voice. But its free beer!):)
Excellent points jargoone. As the author of Voodoo Voice I feel I can clarify one point, however...
You mention that recognition would probably be limited to about 50 commands, and that this would rule out selecting individual songs in the playlist because this would be too much like dictation.
My plugin allows users to select individual songs. In your song example above, the user would have to say: "computer, switch to four smashing pumpkins melloncholy and infinite sadness disc one zero mp three". This is obviously very cumbersome. BUT it would work, and you can always rename your songs to make them easier to speak. I have a friend using my plugin with a 2500 song playlist, some of them as complicated as that. It recognizes those names with a very high (80-90%) accuracy. In fact, it often has _more_ trouble recognizing simpler song titles because there are fewer distinct syllables. The key is that there are only 2,500 or so things that the the speech engine has to look for. With dication, there are, say, 50,00 words, and those can be combined into nearly infite combinations.
Dictation is exponentially more complicated than anything reasonable that a voice enabled MP3 player would need to do. Hence the reason it works so well (even in a noisy car) without training.
This is less of a problem than you might think. With a good uni-directional microphone and the input volume adjusted to an appropriate (low) level, yes, you have to speak over the music. But its not even as bad as yelling something to the person in the passengers seat. I have a Dodge Ram Charger (read: big old, noisy truck) and even with the music cranked my voice recognition plugin (linked above) is quite good at picking up what I say.
Hehe.. I know mean on the K.I.T.T. thing. Perhaps the first thing I did when I got my Voodoo Voice plugin to a semi-working state was implement a personality system that would allow WAV file responses to different commands. And then I created a K.I.T.T. personality. Check it out at the URL above. Just remember, its way early beta, and there's almost no documentation. (working on it!)
The URL above is a link to another solution for controlling an MP3 player with voice commands, called Voodoo Voice. This one is a Winamp plugin and uses Microsoft's SAPI and engine. I've used it in my (very noisy) car with a lot of success. The key is having a high quality uni-directional microphone. It requires very little (if any) training and supports "personalities", which are like voice skins. Although it seems to be working fairly well, it should be considered BETA.
For those of you that doubt the viability of voice control because of your experiences with NaturallySpeaking or ViaVoice, read this: There is a _huge_ difference between dictation and command and control. Voodoo Voice, and almost certainly the original solution mentioned in the article, use command and control. I can almost guarantee you will be surprised at just how good command and control can be with no training at all.
Slashdot Mirror
Unfortunately, it does require you to say the entire, exact name of the song. I've been trying to come up with a way around this, but haven't had much success. For now the best solution is to give your songs easier names.
Voodoo Voice works through Microsoft's Speech API which simply defines an interface between voice applications and engines. I've done most of my testing with Microsoft's freely available engine, but if you have NaturallySpeaking or ViaVoice it will use that engine instead and you will probably get better results.
(Yes, shameless plug, since I'm the author of Voodoo Voice. But its free beer!) :)
With a good uni-directional microphone, the computer won't even hear it. Unless of course you put the microphone 3 inches from the speakers. :)
You mention that recognition would probably be limited to about 50 commands, and that this would rule out selecting individual songs in the playlist because this would be too much like dictation.
My plugin allows users to select individual songs. In your song example above, the user would have to say: "computer, switch to four smashing pumpkins melloncholy and infinite sadness disc one zero mp three". This is obviously very cumbersome. BUT it would work, and you can always rename your songs to make them easier to speak. I have a friend using my plugin with a 2500 song playlist, some of them as complicated as that. It recognizes those names with a very high (80-90%) accuracy. In fact, it often has _more_ trouble recognizing simpler song titles because there are fewer distinct syllables. The key is that there are only 2,500 or so things that the the speech engine has to look for. With dication, there are, say, 50,00 words, and those can be combined into nearly infite combinations.
Dictation is exponentially more complicated than anything reasonable that a voice enabled MP3 player would need to do. Hence the reason it works so well (even in a noisy car) without training.
This is less of a problem than you might think. With a good uni-directional microphone and the input volume adjusted to an appropriate (low) level, yes, you have to speak over the music. But its not even as bad as yelling something to the person in the passengers seat. I have a Dodge Ram Charger (read: big old, noisy truck) and even with the music cranked my voice recognition plugin (linked above) is quite good at picking up what I say.
Hehe.. I know mean on the K.I.T.T. thing. Perhaps the first thing I did when I got my Voodoo Voice plugin to a semi-working state was implement a personality system that would allow WAV file responses to different commands. And then I created a K.I.T.T. personality. Check it out at the URL above. Just remember, its way early beta, and there's almost no documentation. (working on it!)
For those of you that doubt the viability of voice control because of your experiences with NaturallySpeaking or ViaVoice, read this: There is a _huge_ difference between dictation and command and control. Voodoo Voice, and almost certainly the original solution mentioned in the article, use command and control. I can almost guarantee you will be surprised at just how good command and control can be with no training at all.