1200574 - Add Pocketsphinx speech model to gaia-l10n/fr

Status: RESOLVED FIXED

(Mozilla Localizations :: fr / French, defect)

Description

[kdavis]

As part of Bug 1178288 the fr language pack will include a speech
recognition model for fr. Thus, this model should thus be included
in the gaia-l10n/fr repository.

Comment 1

[kdavis]

I have the models prepared and they simply need to be committed to the repository.
The suggested directory layout is as follows: 

fr
├── apps
│   ├── bluetooth
│   ├── ...
├── shared
│   ├── branding
│   ├── ...
└── speech-data
    ├── feat.params
    ├── fr.dic
    ├── fr.dic.dmp
    ├── mdef
    ├── means
    ├── mixture_weights
    ├── noisedict
    ├── sendump
    ├── transition_matrices
    └── variances

Comment 2

[Théo Chevalier]

Hi Kelly,

Feel free to attach a .zip or a patch to this bug and I can commit that for you.

Comment 3

[kdavis]

Attachment: Addition of Pocketsphinx speech model

Addition of Pocketsphinx speech model

Comment 4

[Théo Chevalier]

Comment on attachment 8655516 [details]
Addition of Pocketsphinx speech model

Thanks, landed in https://hg.mozilla.org/gaia-l10n/fr/rev/d48187deb314

Rubber-stamping it more than actually reviewing it ;)

Btw, where is the dictionary coming from? Pocketsphinx?

Any chance we can use http://www.dicollecte.org/download.php?prj=fr in the future? This is what we use and keep up-to-date on Desktop: http://hg.mozilla.org/l10n-central/fr/file/default/extensions/spellcheck/hunspell/

Comment 5

[kdavis]

(In reply to Théo Chevalier (ooo 08/24 - 09/01) [:tchevalier] from comment #4)
> Comment on attachment 8655516 [details]
> Addition of Pocketsphinx speech model
> 
> Thanks, landed in https://hg.mozilla.org/gaia-l10n/fr/rev/d48187deb314

Thanks!

> Btw, where is the dictionary coming from? Pocketsphinx?

Yes.
 
> Any chance we can use http://www.dicollecte.org/download.php?prj=fr in the
> future? This is what we use and keep up-to-date on Desktop:
> http://hg.mozilla.org/l10n-central/fr/file/default/extensions/spellcheck/
> hunspell/

In future we are going to move away from the "dictionary", it's actually more
a phonetic dictionary giving each word's pronunciation, to a weighted finite
state transducer that is able to spell a word phonetically from its written
form. (The dic.dmp file is actually just such a transducer model.)

If http://www.dicollecte.org/download.php?prj=fr contains phonetic dictionary
we could use it in future to train the transducer.

Comment 6

[Théo Chevalier]

(In reply to kdavis from comment #5)
> (In reply to Théo Chevalier (ooo 08/24 - 09/01) [:tchevalier] from comment
> #4)
> > Comment on attachment 8655516 [details]
> > Addition of Pocketsphinx speech model
> > 
> > Thanks, landed in https://hg.mozilla.org/gaia-l10n/fr/rev/d48187deb314
> 
> Thanks!
> 
> > Btw, where is the dictionary coming from? Pocketsphinx?
> 
> Yes.
>  
> > Any chance we can use http://www.dicollecte.org/download.php?prj=fr in the
> > future? This is what we use and keep up-to-date on Desktop:
> > http://hg.mozilla.org/l10n-central/fr/file/default/extensions/spellcheck/
> > hunspell/
> 
> In future we are going to move away from the "dictionary", it's actually more
> a phonetic dictionary giving each word's pronunciation, to a weighted finite
> state transducer that is able to spell a word phonetically from its written
> form. (The dic.dmp file is actually just such a transducer model.)
> 
> If http://www.dicollecte.org/download.php?prj=fr contains phonetic dictionary
> we could use it in future to train the transducer.

oh, ok, I thought it was a regular dictionary. I don't think Dicollect has a phonetic dictionary yet. Thanks for the clarification!

Closing the bug since the patch has landed.