Can Rockbox be made to turn a Sansa into a voice recognition mp3 player?

[cocopugg]

This may be a stupid question, but since the Sansa comes with a voice recorder, can it be technically possible for Rockbox software to allow for voice commands on the Sansa player? In other words, can it be set up so when Rockbox boots up, it boots up to the voice recorder (but as a voice recognition program) and instead of physically having to press menu buttons to navigate where we want to, we can say something like "Play Madonna Folder" and it would start playing back your Madonna folder. Like I said, it's probably a dumb question, but I was wondering since the Sansa does have a voice recorder, and the Rockbox crew is so great at what they've done with the Rockbox software so far, it might be a question someone here could answer.

[psycho_maniac]

you can set up rockbox to start in the recording screen but it does not have a "voice recognition" feature.

[cocopugg]

you can set up rockbox to start in the recording screen but it does not have a "voice recognition" feature.

I understand that. I was wondering if such a feature could actually be written within Rockbox. Actually since Rockbox has features to allow blind people to use their audio files to guide them, maybe integrating a voice recognition feature into the program might be something the Rockbox crew might be interested in looking into. Though I'm not sure how much space such a program could take up. Anyway, I was just playing with that thought. You know the day WILL come when someone develops such an mp3 player, if they haven't already.

[saratoga]

This may be a stupid question, but since the Sansa comes with a voice recorder, can it be technically possible for Rockbox software to allow for voice commands on the Sansa player?

Yes its technically possible.

 

[safetydan]

Actual voice recognition (as in you could voice recognition to type like DragonDictate or similar) is not likely to be doable on a typical MP3 player. There's just not enough CPU power. You might be able to get away with sample matching like some mobile phones have where there's a limited set of commands recognised.

[pondlife]

As a start, maybe just monitor the mic input and allow any sharp transient (say, hand clapping) trigger a "next track".  

Of course, no matter what calibration you can do, playback at loud volume is quite likely to result in unwanted skipping, but you'd be on the first step towards voice input.

pondlife

[dd]

heya,

I don't know whether it's possible to use the source code of PocketSphinx [1]. It is a speech recognition program specifically designed for handhelds running linux (including code also for ARM-processors). It's a fork of Sphinx2 which uses HMMs (Hidden Markov Models) with semi-continuous output probability density functions. License is BSD-like.

It seems to be fast enough for our problem (but i don't know anything about ARM-processors…). I think, we will have another problem about speech recognition: The surrounding noise. My sansa has a microphone which can record people talking 10 meters apart…  Maybe one could control this by »loudness-equalizing«⁇
Another problem could be the dialect especially on Artist's names like Madonna… Native English/Italian/German speakers would all spell it out in another way. This problem could be solved by generating  language specific dictionarys.
So, if anyone with knowledge on ARM-processor-power could check it out, we'd propably get a neat thing called voice input.

dd

[1]: http://www.speech.cs.cmu.edu/pocketsphinx/

[zajacattack]

Well, in my opinion, it wouldn't be worth it. To make optimized AND accurate voice recognition in rockbox is an oxymoron. The processors are no where near powerful enough.

[GodEater]

If the product is designed for embedded devices, which it says it is - it should be possible to run it on at least SOME Rockbox targets.