Forums

methodologies for speech recog

Started by Paresh Chopdekar September 22, 2002
Hi

I am a new member of this society. I am a final year B.E. student from
Mumbai. Formy final year project i have selected Speech Recognition as my topic.

I have a few basic questions :-

1 Which is the best method for feature extraction, cepstrum,LPC,PLP,Mel
Cepstral
2 I am planning to use a ANN for phoneme recognition, is it feasible & will
it provide proper recognition?
3 How do I construct a word-phoneme database? How do i convert a word to its
phonetic representation?

I am a beginner in this field, so please provide detailed info or liks to sites
that provide such info

Thanx in advance
Paresh C



Hi Paresh,
In MY opinion, these are the answers to your questions:

1. In case of clean speech, MFCC (Mel cepstrum) are the most reliable features.
However, in noisy conditions, PLP-Rasta are supposed to give better performance.

2. ANN has been widely used for phoneme recognition. Ofcourse it is feasible and
gives best performance if trained properly. So you should have a good training
databse on which you can train your ANN. It also depends on what application you
are targetting? Is it continuous speech recognition? or is it just isolated word
recognition?

3. For phonetic transcription...i dont have much idea...however there are some
databses distributed by LDC, which are accompanied by the phonetic
transcriptions of the word. I think, a visit to CSLR website should also help. goodluck,
--g2

"Paresh Chopdekar"<> wrote:
Hi

I am a new member of this society. I am a final year B.E. student from
Mumbai. Formy final year project i have selected Speech Recognition as my topic.

I have a few basic questions :-

1 Which is the best method for feature extraction, cepstrum,LPC,PLP,Mel
Cepstral
2 I am planning to use a ANN for phoneme recognition, is it feasible & will
it provide proper recognition?
3 How do I construct a word-phoneme database? How do i convert a word to its
phonetic representation?

I am a beginner in this field, so please provide detailed info or liks to sites
that provide such info

Thanx in advance
Paresh C

_____________________________________
Note: If you do a simple "reply" with your email client, only the author of this
message will receive your answer. You need to do a "reply all" if you want your
answer to be distributed to the entire group.

_____________________________________
About this discussion group:

To Join:

To Post:

To Leave:

Archives: http://www.yahoogroups.com/group/speech-recognition

Other DSP-Related Groups: http://www.dsprelated.com

">http://docs.yahoo.com/info/terms/ Get Your Private, Free E-mail from Indiatimes at http://email.indiatimes.com

Buy Music, Video, CD-ROM, Audio-Books and Music Accessories from
http://www.planetm.co.in

Change the way you talk. Indiatimes presents Valufon, Your PC to Phone service
with clear voice at rates far less than the normal ISD rates. Go to
http://www.valufon.indiatimes.com. Choose your plan. BUY NOW.


hi
thankz for the quik and clear reply...at present i am
working on isolated word recognition using
wavelets...i will definetly let u know when i come to
some conclusion...
and regarding ANN, i dont have any idea of that
concept...as per ur suggetion i will go through the
concept first and then i will try to use HMM with
ANN...
can u ask ur friends who have done on wavelets to mail
me the procedure they have followed for speech
recognition...i would like to go in depth of wavelets
to come to a cofirmed conclusion...
thanking u
with regards
usha
--- jitendraajmera <>
wrote:
> Hi,
> I know a little bit about wavelets....but havent
> tried it for speech recognition. HOwever, some
> colleagues of mine and also other researchers have
> tried it and did not find helpful (for no clear
> reasons).
>
> However, I am keenly interested in knowing the
> results of your research.....using wavelets.
>
> Now, about phoneme recognition thingy by
> Paresh....if you have a training database with
> phonetic transcription, ANN is the best choice. At
> the time of recognition/decoding, one has to use
> HMMs....there is no other choice. However, in
> absense of phonetic transcription for the training
> data, using HMMs are inevitable for training too.
>
> If I am wrong at some point, I would be grateful for
> your corrections.
>
> best of luck for wavelets,
> cheers,
> --g2 > PS: I think, you forgot to reply your last mail to
> the whole group...so the mails seem to be floating
> between two of us.
> "usha devi"<> wrote:
> hi
> If u want to go for phoneme recognition,in my view
> the
> best option would be HMM(hidden markov model)....
>
> At present i am working on speech recognition(number
> and isolated word recognition) using wavelets...can
> any body suggest me the correct choice of wavelet
> that
> can be used for speech recognition....
>
> I would like to build up my own filter co-efficients
> for the subband coding...i need suggestions in this
> regard too...
>
> with regards
> usha
> --- jitendraajmera <>
> wrote:
> > Hi Paresh,
> > In MY opinion, these are the answers to your
> > questions:
> >
> > 1. In case of clean speech, MFCC (Mel cepstrum)
> are
> > the most reliable features. However, in noisy
> > conditions, PLP-Rasta are supposed to give better
> > performance.
> >
> > 2. ANN has been widely used for phoneme
> recognition.
> > Ofcourse it is feasible and gives best performance
> > if trained properly. So you should have a good
> > training databse on which you can train your ANN.
> It
> > also depends on what application you are
> targetting?
> > Is it continuous speech recognition? or is it just
> > isolated word recognition?
> >
> > 3. For phonetic transcription...i dont have much
> > idea...however there are some databses distributed
> > by LDC, which are accompanied by the phonetic
> > transcriptions of the word. I think, a visit to
> CSLR
> > website should also help.
> >
> >
> > goodluck,
> > --g2
> >
> > "Paresh Chopdekar"<> wrote:
> > Hi
> >
> > I am a new member of this society. I am a
> final
> > year B.E. student from Mumbai. Formy final year
> > project i have selected Speech Recognition as my
> > topic.
> >
> > I have a few basic questions :-
> >
> > 1 Which is the best method for feature
> > extraction, cepstrum,LPC,PLP,Mel Cepstral
> > 2 I am planning to use a ANN for phoneme
> > recognition, is it feasible & will it provide
> proper
> > recognition?
> > 3 How do I construct a word-phoneme database?
> How
> > do i convert a word to its phonetic
> representation?
> >
> > I am a beginner in this field, so please provide
> > detailed info or liks to sites that provide such
> > info
> >
> > Thanx in advance
> > Paresh C
> >
> >
> >
> >
> > _____________________________________
> > Note: If you do a simple "reply" with your email
> > client, only the author of this message will
> receive
> > your answer. You need to do a "reply all" if you
> > want your answer to be distributed to the entire
> > group.
> >
> > _____________________________________
> > About this discussion group:
> >
> > To Join:
> >
> >
> > To Post:
> >
> > To Leave:
> >
> >
> > Archives:
> >
> http://www.yahoogroups.com/group/speech-recognition
> >
> > Other DSP-Related Groups:
> http://www.dsprelated.com
> >
> > ">http://docs.yahoo.com/info/terms/
> >
> >
> > Get Your Private, Free E-mail from Indiatimes at
> > http://email.indiatimes.com
> >
> > Buy Music, Video, CD-ROM, Audio-Books and Music
> > Accessories from http://www.planetm.co.in
> >
> > Change the way you talk. Indiatimes presents
> > Valufon, Your PC to Phone service with clear voice
> > at rates far less than the normal ISD rates. Go to
> > http://www.valufon.indiatimes.com. Choose your
> plan.
> > BUY NOW.
> >
> >
> > __________________________________________________
>
> http://email.indiatimes.com
>
> Buy Music, Video, CD-ROM, Audio-Books and Music
> Accessories from http://www.planetm.co.in
>
> Change the way you talk. Indiatimes presents
> Valufon, Your PC to Phone service with clear voice
> at rates far less than the normal ISD rates. Go to
> http://www.valufon.indiatimes.com. Choose your plan.
> BUY NOW.
>


__________________________________________________