My 6 year old daughter has due to ion channel issues balance iand speech issues. When she speaks to others the issue is that they cannot understand her.
Because she drops letters or speaks instead of a specific letter in the word another letter. It's an own language that we as "trained" parents understand.
I want give her a tool that translates her words into understandable language.(of course we try to improve her speech with intensive therapies)
The idea is that she gets a tiny recorder (edic-mini tiny) attached on her shirt and we analyse everday the words that she speaks for kaldi training.
Can anybody give me a hint how i should proceed?
Sould we use just her words for training or mix it with another speech corpus?
Should the neuronal betworks be used?
Should we use another recorder instead of the edic?
Does somebody already build this with Kaldi?
Thank you in advance for any help in this project.
Best regards,
Joseph
Last edit: Joseph D 2015-06-01
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I think it would be too much work to build an application like this
for just one person. Even if you could get reasonable accuracy, it
would require a huge amount of work to build a usable system around
it, and you would need a team with experience in both speech
recognition and UI design.
Dan
My 6 year old daughter has due to ion channel issues invrain cells balance
issues and speech issues. When she speaks to others the issue is that they
cannot understand her.
Because she drops letters or speaks instead a specigic letter in the word
another letter. It's an iwn language that we as "trainef" parents
understand.
I want give her a tool that translates her words into understandable
language.(of course we try to improve her speech with intensive therapies)
The idea is that she gets a tiny recorder (edic-mini tiny) attached on her
shirt and we analyse everday the words that she speaks for kaldi training.
Can anybody give me a hint how i should proceed?
Sould we use just her words for training or mix it with another speech
corpus?
Should the neuronal betworks be used?
Should we use another recorder instead of the edic?
Does somebody already build this with Kaldi?
Thank you in advance for any help in this project.
You could definitely try that. Diarization in Kaldi isn't very
developed at the moment.
You will also have to build a user interface for the manual transcription.
Dan
Dan. Yes i know it will be a lot of effort but i'll have to do it. Thank you
for your support so far (Also outside of the forum)
So i'll do i step by step.
The first think (in parallel to setting up a kaldi installation) for me
would be the split of the recording. (e.g. 6 hours)
What do you think about using
"http://lium3.univ-lemans.fr/diarization/doku.php"?
Or how would you split the data and generate an individual speech corpus?
Hi,
My 6 year old daughter has due to ion channel issues balance iand speech issues. When she speaks to others the issue is that they cannot understand her.
Because she drops letters or speaks instead of a specific letter in the word another letter. It's an own language that we as "trained" parents understand.
I want give her a tool that translates her words into understandable language.(of course we try to improve her speech with intensive therapies)
The idea is that she gets a tiny recorder (edic-mini tiny) attached on her shirt and we analyse everday the words that she speaks for kaldi training.
Can anybody give me a hint how i should proceed?
Sould we use just her words for training or mix it with another speech corpus?
Should the neuronal betworks be used?
Should we use another recorder instead of the edic?
Does somebody already build this with Kaldi?
Thank you in advance for any help in this project.
Best regards,
Joseph
Last edit: Joseph D 2015-06-01
I think it would be too much work to build an application like this
for just one person. Even if you could get reasonable accuracy, it
would require a huge amount of work to build a usable system around
it, and you would need a team with experience in both speech
recognition and UI design.
Dan
On Mon, Jun 1, 2015 at 4:23 PM, Joseph Dirmeyer josephd112@users.sf.net wrote:
I know of a company that is developing a commercial product like you are describing.
http://www.talkitt.com/
hope this helps.
Michael
@Michael Thank you. Yes i know this company. But currently i only see a lot of marketing acitivities.
Last edit: Joseph D 2015-06-02
Dan. Yes i know it will be a lot of effort but i'll have to do it. Thank you for your support so far (Also outside of the forum)
So i'll do i step by step.
The first think (in parallel to setting up a kaldi installation) for me would be the split of the recording. (e.g. 6 hours)
What do you think about using http://lium3.univ-lemans.fr/diarization/doku.php ?
or http://shout-toolkit.sourceforge.net/index.html
for the first task
Or how would you split the data and generate an individual speech corpus?
Last edit: Joseph D 2015-06-02
You could definitely try that. Diarization in Kaldi isn't very
developed at the moment.
You will also have to build a user interface for the manual transcription.
Dan
On Tue, Jun 2, 2015 at 5:22 AM, Joseph D josephd112@users.sf.net wrote: