[Top][All Lists]

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

Re: [Accessibility] Call to Arms

From: Eric S. Johansson
Subject: Re: [Accessibility] Call to Arms
Date: Wed, 28 Jul 2010 12:44:03 -0400
User-agent: Mozilla/5.0 (Windows; U; Windows NT 6.1; en-US; rv: Gecko/20100713 Thunderbird/3.1.1

 On 7/28/2010 11:59 AM, Christian Hofstader wrote:
cdh: I have an ever growing file of items related to voice recognition. I think one of the coolest ideas came from rms who suggested that we send people a corpus and have them record themselves reading it. We then use these recordings to train speech recognition engines and see how the results work out.

have you looked up how different recording environments, microphones and sound systems affect the quality of the audio? From what I understand, want it extremely consistent extremely high-quality audio for your corpus and then degrade it in a deterministic way to work with different kinds of sound systems (i.e. Bluetooth). For example, you don't want someone reading the corpus on the blue tooth ear bud in downtown Cambridge. You do want someone using a high-quality microphone like a Sennheiser in a well insulated/acoustically isolated room with cloth walls using a 24-bit audio device.

cdh: I think we need to learn some details on how widely the range of accents we can accept varies. For instance, do most people need to speak with a relatively "standard" US accent or do our friends up in Harlem and down in Mississippi need to be excluded?

or the Indian-American, Australia, British etc. accents. this is a solved problem and solutions have existed in the public knowledgebase since about 2003. We'll probably have more success finding the information you need at a library that online

cdh: I'll be in this motel until Friday when we start our drive north to our Cambridge residence and I'll be online a lot. Can someone volunteer to help collaborate with me on writing a two or three page project proposal that will help list the tasks related to speech recognition, skills of volunteers, existing components and help me find a person to lead the accessibility speech-recognition project? We will, of course, put this document out for comment on this list to refine it but, as I said in my original call to arms mail, it's time to get hacking.

it would have been nice to help but I have no tools with which I can contribute. I've started my search for a nonfree tool chain (editor etc.) so maybe I'll be able to contribute sometime in the future.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]