speech-to-text module???

Robofreak

Hello all,

Just wondering... I've heard of Text-to-Speech modules, such as the EmicTTS, but has anyone seen a SPEECH-to-TEXT module before? That would be fun to put on a robot...

Thanks for any input,
Robofreak

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Austin Bowen,

Robo-freak.com

"One must watch out for mechanics. They start out with a sewing machine, and end up with the atomic bomb" - A quote from someone that I saw on Addall.com

MSDTech

If you find a seperate module, let me know. I found the finger spelling robot in the customer applications fascinating and would love to duplicate that with a speech to text module. I am looking at the Microsoft Robotics Studio and linking that with the speech to text capabilites that are built into some of the Office products.
In my position, working at a school for the deaf, this would be very usefull. My ASL is not good and I can use any help I can get.

Mike Green

Speech to text is much much harder. There are systems that run on PCs and Macs that require hundreds of megabytes of memory to work at a reasonable speed and they're not close to perfect. Dragon Naturally Speaking is one system for PCs and MacSpeech is one for Macs.

Robofreak

Would there happen to be a module that you can program through serial comunications on startup (such as with a propeller) where it'll give an identfication value if it hears a certain word (which it would be told on startup)? You know, it listens for certain words and gives an output when it hears it? For instance, a program where it scans for the word "ROBOT" then esponds with the EmicTTS saying, "Yes?" Then you could tell it to do some pre-programmed stuff...

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Austin Bowen,

Robo-freak.com

"One must watch out for mechanics. They start out with a sewing machine, and end up with the atomic bomb" - A quote from someone that I saw on Addall.com

Mike Green

Here's a link to one manufacturer: www.sensoryinc.com/products/modules_and_boards.html. This isn't simple to use though. There have been some simple schemes for voice recognition of a small number of words (10-15) using simple hardware (zero crossing detector for primary frequency detection) and simple software (ran on an Apple II). It wasn't very good though. The software looked for quiet periods between words, then normalized the sound by determining the dominant frequency over time and stretching or compressing the data to fit a specific size buffer. It would compare the frequency pattern to one of several saved patterns and report the closest match or a failure.