← Back to team overview

ubuntu-phone team mailing list archive

Re: Voice Recognition

 

Hi All,

The current system uses the open source library pocketsphinx (
http://cmusphinx.sourceforge.net/) - I believe this is the current "best"
open source voice recognition tool.

Unfortunately this library is not able to perform dictation with an
acceptable level of accuracy. I don't know of any open source voice tools
that are.

It is applicable to "command and control" situations, like we use it for in
HUD. It can work with either

   - A language model. Built from a corpus of text, e.g. the Wall Street
   Journal. This basically gives it information the probabilities different
   words being together in speech. Theoretically this can be used to recognise
   more natural commands, but I found its accuracy to be very poor.
   - A finite state grammar - this is what we use in HUD. Basically it's a
   simple state machine that defines the commands that we get from the various
   programs running. This has relatively high accuracy, but has the limitation
   of only recognising a fixed set of inputs.

It is worth noting that a considerable amount of effort was expended
training the acoustic model (the part that actually identifies what
phonemes the user is saying) that is used currently. It was trained against
the Voxforge international english audio corpus (http://voxforge.org/).

Much of the academic work done in this area is funded by the likes of
Microsoft. In these situations a pretty strict "closed licence" is usually
applied. Unfortunately what this means practically, is that the only way
you'll get dictation at the moment is with a closed-source tool...

I hope this information is useful!

Cheers,
    Pete


On Thu, Jun 27, 2013 at 9:22 AM, Zisu Andrei <matzipan@xxxxxxxxx> wrote:

> I wrote a little something for this occasion:
> http://zisuandrei.webmonsters.ro/design-in-open-source-unity-and-dash/ there
> are some sections called "HUD" and "Dash" and "HUD voice interaction". Some
> of my propositions are obsolete after 3 months from writing, some are
> outright stupid, but bear with my article:D
>
> I tried desperately to get on the unity-design mailing list but it seems
> there is no one in charge of approving those mailing list acceptance
> requests (https://launchpad.net/~unity-design/+members#proposed 44
> membership requests for over 3-4 months). If there is any Canonical
> employee that could help me with approving my membership for unity-design I
> would greatly appreciate it.
>
> Zisu Andrei
>
>
> On 27 June 2013 03:38, Josh Leverette <coder543@xxxxxxxxx> wrote:
>
>> there seems to be a basic system for voice recognition, designed to
>> control the currently open app based on a limited command vocabulary. I'm
>> curious if there are any plans for a Siri-like assistant, maybe for the
>> 14.04 release? and how accurate is the dictation software? is it good
>> enough for straight text composition like Joseph was talking about? I know
>> it's fully offline, which is a tremendous advantage and disadvantage over
>> the competition.
>>
>> Sincerely,
>> Josh
>> On Jun 26, 2013 9:35 PM, "Joseph Bylund" <joseph.bylund@xxxxxxxxx> wrote:
>>
>>> Has there been any discussion of voice recognition, both to compose
>>> texts/emails and to perform actions, like "open the music player"? I find
>>> this feature quite useful on my current phone.
>>>
>>> -Joe
>>>
>>> --
>>> Mailing list: https://launchpad.net/~ubuntu-**phone<https://launchpad.net/~ubuntu-phone>
>>> Post to     : ubuntu-phone@lists.launchpad.**net<ubuntu-phone@xxxxxxxxxxxxxxxxxxx>
>>> Unsubscribe : https://launchpad.net/~ubuntu-**phone<https://launchpad.net/~ubuntu-phone>
>>> More help   : https://help.launchpad.net/**ListHelp<https://help.launchpad.net/ListHelp>
>>>
>>
>> --
>> Mailing list: https://launchpad.net/~ubuntu-phone
>> Post to     : ubuntu-phone@xxxxxxxxxxxxxxxxxxx
>> Unsubscribe : https://launchpad.net/~ubuntu-phone
>> More help   : https://help.launchpad.net/ListHelp
>>
>>
>
> --
> Mailing list: https://launchpad.net/~ubuntu-phone
> Post to     : ubuntu-phone@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~ubuntu-phone
> More help   : https://help.launchpad.net/ListHelp
>
>

Follow ups

References