Click here to register.

Acoustic Model Discussions

Flat
training with radio/tv sounds
User: ubanov
Date: 11/18/2008 5:23 pm
Views: 2951
Rating: 2

Hi,

Does anybody know if will be legal to use the sounds from a Radio or TV broadcast in the process of training a language. I suppose that this sounds could not be uploaded to voxforge, but could any person use this sounds as the source to the training (making the transcription, or even using comercial speech to text software to make the transcription).

I'm thinking in some languages (like spanish) that don't have any voice recognition model and they have too little quantity of voices in voxforge.

Could it be posible this?

Thanks in advace

--- (Edited on 11/18/2008 5:23 pm [GMT-0600] by ubanov) ---

Re: training with radio/tv sounds
User: kmaclean
Date: 11/18/2008 7:56 pm
Views: 1222
Rating: 2

HI ubanov,

>Does anybody know if will be legal to use the sounds from a Radio or TV

>broadcast in the process of training a language.

Creating an acoustic model from Radio or TV would likely be considered a derivative work, and thus be covered under the Copyright of the creator of the work.

There is an argument that since you cannot reverse engineer the acoustic model back to the original Radio or TV program, that this is not really a derivative work (Google's Google Web 1T 5-gram Corpus is a good example of this in a language model context).  As opposed to the case where including a recorded section of of someone else's song in your own music would clearly be an illegal use of a derivative of the original work... 

I don't know the answer to this, and don't have the resources (i.e. money) to find out (i.e. if someone decides they disagree with us and want to take us to court...).  Therefore, we've been taking a conservative approach with respect to source speech audio Copyright issues, and thus try to make sure that we have clear authority to use/distribute any speech audio used creating the Voxforge acoustic models.

In addition, since we want to create GPL acoustic models, we need to be able to distribute the source audio in addition to the derivative acoustic models, and the unauthorized distribution of the original source radio or TV programs would clearly be in breach of Copyright.

Ken

 

 

--- (Edited on 11/18/2008 8:56 pm [GMT-0500] by kmaclean) ---

PreviousNext