To LUGNET HomepageTo LUGNET News HomepageTo LUGNET Guide Homepage
 Help on Searching
 
Post new message to lugnet.roboticsOpen lugnet.robotics in your NNTP NewsreaderTo LUGNET News Traffic PageSign In (Members)
 Robotics / 12005
12004  |  12006
Subject: 
An interesting diversion - speech!
Newsgroups: 
lugnet.robotics
Date: 
Mon, 7 Aug 2000 15:26:09 GMT
Original-From: 
John Barnes <barnes@sensors&spamless&.com>
Viewed: 
734 times
  
I spent this weekend interfacing the Sensory Voice Direct 364
speech recognition device to the RCX as a sensor.

Technically, it wasn't that hard. The device has 8 outputs which
switch from a logic 0 to a logic 1 as each of 8 words are detected.

Then it leaves the 8th output high and repeats the process for the
next 7 words on the remaining 7 outputs.

I arranged for each output to have a unique resistor value attached
to it to provide equally spaced readings at the RCX end when the
connection was configured as passive/raw. The 8th output it given
a special half value so that it can be detected by itself for word 8
and in conjunction with the other 7 outputs for words 9 - 15.

I powered the module from 4 x AA NiMH batteries because I noted
that the capacitors on the board are only rated at 6.4 volts so normal
1.5 volt cells, when new, could stress that rating. The unit draws
about 40mA most of the time, so rechargeables are a good idea
anyhow.

Eventually, I revised the interface connection to use 8 MOSFETs
(2N7000) to drive the resistor interface to the RCX which removed
an unacceptable dependence on the state of charge of the batteries
from the RCX's readings.

In use, its Ok. Not great, but Ok. It does continuous recognition so
it requires a keyword to attract its attention. Then it processes the
next word (which may in fact be a phrase) to decide which output
to enable.

So, with a keyword like "Robot",

You train it for things like;

Robot - Stop
Robot - Go
Robot - Turn left
Robot - Turn right

I say Ok, because it's not quite as good as the literature might lead you
to believe. In fact, it cannot distinguish between "left" and "right",  (a
common problem with people also but for different reasons :) so I biased
the training to leFFT and Ri, emphasizing the "FFT" on left and the "I" vowel
in right. Then it works fine.

The only thing I changed, apart from adding the RCX interface parts was
the speaker you get in Sensory's SDK which is a little large. I used a
smaller speaker, a flat one from Radio Shack, and also an op-amp on the
microphone input so it will work at greater range. The whole thing is
installed
in a 10 x 10 x 4 big brick.

For $50, I just had to know if it would work. On balance, it's an interesting
demo, but not really that important. It chews up an RCX input and doesn't
work quite as well as I had hoped.

email me if you'd like more details.

JB

The device is available from JDR. Make sure you order the 364 and not the
earlier version which doesn't have continuous recognition.



Message has 2 Replies:
  Re: An interesting diversion - speech!
 
(...) Just one word: GREAT! [big snip of many technical details] (...) I'll do. And please continue to surprise us with your incredible devices. Ciao Mario (24 years ago, 7-Aug-00, to lugnet.robotics)
  Re: An interesting diversion - speech!
 
I just received my All Electronics Corporation Summer catalog and they have the 364 on page 3 for 49.95 as well... it lists a vocabulary of 60 word/phrases so I would imagine that it is outputting a binary match code for the phrases. (URL) will hook (...) (24 years ago, 8-Aug-00, to lugnet.robotics)

5 Messages in This Thread:


Entire Thread on One Page:
Nested:  All | Brief | Compact | Dots
Linear:  All | Brief | Compact
    

Custom Search

©2005 LUGNET. All rights reserved. - hosted by steinbruch.info GbR