[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [AUDITORY] Phoneme database

Dear Giampiero,
thank you for your reply. I would rather like to have the second case of course. And this is what I had in mind what you've suggested, in case that I cannot find an already isolated one. But there are still those two databases (Oxford and USF). I don't know how they made the recordings. In particular the one from USF seems to have visual info as well.

Best Regards

On Thu, Oct 31, 2013 at 9:58 AM, Giampiero Salvi <giampi@xxxxxx> wrote:
Dear Kamil,
what do you mean exactly with "isolated phoneme database"? Do you mean
phonemes spoken in isolation, or a database with files each containing
a phone extracted from continuous speech recordings?

In the first case, the models you train won't be of much use for ASR:
due to coarticulation, the actual sounds (phones) you will find in
speech will differ greatly from your isolated phoneme recordings.

In the second case, you could generate such database from TIMIT with a script.

Best regards

On 31 October 2013 09:39, Kamil Adiloglu <kamil.adiloglu@xxxxxxxxx> wrote:
> Dear list,
> I am looking for a phoneme database to train a supervised machine learning
> method for ASR. There is the well-known TIMIT database, which contains the
> phoneme labels for each recording, but an isolated phoneme database would be
> much better for me. I also found some websites mentioning  the Oxford
> Acoustic Phonetic Database or the University of South Florida Audiovisual
> Phone Database, but no clue about how to find the databases themselves. Any
> help would be highly appreciated.
> Best Regards
> Kamil Adiloglu

Giampiero Salvi
Associate Professor
KTH, School of Computer Science and Communication