You has my skype
British english
|
As some of you might know, I'm working on a cross-platform speech-recognition 'plugin' for controlling vim-editor environments.
I've got my design laid out, but once it's done, I will need your help!
The speech recognition works with neural networks. Neural networks are like living creatures, they need to be trained and tamed!
These trained networks are called "Acoustic models".
The training consists of reading a file with vim commands out loud and sending the audio to the neural network. It will then try to learn how to map the vim command to your voice.
The problem is, I have a heavy accent, so when I train the neural network, it will probably not understand a native speaker.
I want to make this application available for everyone so here is the plan:
I want 1 acoustic model that is trained by multiple accents (British, scottisch, finnish-english, american, canadian, dutch-english ..) etc. This acoustic model is like a one-size-fits-all model. It will work for most people.
Then I want to train an acoustic model per accent. If your'e dutch, you take the dutch-english acoustic model, if you're german you take the german-english model etc...
This way, everyone around the world can use my application. The one-size-fits-all is actually a contingency plan for when 'your' acoustic model is not available yet, and you want to have something generic that probably works.
So yeah, if you're interested, either pm me or write a response to this post in the following format:
skypename
first language
second language
third language
Based on this information I'll choose the appropiate test subjects.
Don't have your hopes up, testing will probably start not any time soon. I expect we can start recording the first acoustic models in autumn (fall)..
I'll probably create some kind of GUI application that will make it really easy for the testers to submit new acoustic models to the database.
When the application is running, anyone can contribute their acoustic models to the database and I'll probably keep updating the one-size-fits-all model as a mean of all uploaded models.
------
I might eventually write a plugin system, so that you can create speech-recognized tasks for other applications as well. The user can then submit a grammar, a model and the handler.
The grammar defines what a user can say in a specific context. The neural network is built around the grammar, and trained with the model. If the neural network detects a 'hit'. it fires off an event (Hey this was said, who can handle this stuff). Then he user can connect handlers to the network, that handle specific sentences, tokens and words. These handlers can then execute the appropriate actions of the command.
You have my Skype
Southern English (USA)
Also, pleaaase let me do the GUI
Dagoberto.lopez 2
English (Florida, USA)
If needed graphics layout I can help
steven.kreitzer
the accent that is on the news in the united states, the normal one, english
You've got mine; I speak with a more neutral American accent but can speak with a New York/New England accent (grew up there).
Last edited by Mister Maggot; 05-09-2013 at 09:11 PM.
koygfx
Central Canadian accent.
« Previous Thread | Next Thread » |
Thread Information |
Users Browsing this ThreadThere are currently 1 users browsing this thread. (0 members and 1 guests) |