Welcome to Australia's only Vintage Radio and Television discussion forums. You are not logged in. Please log in below, apply for an account or retrieve your password.
Australian Vintage Radio Forums
  Home  ·  About Us  ·  Discussion Forums  ·  Glossary  ·  Outside Links  ·  Policies  ·  Services Directory  ·  Safety Warnings  ·  Tutorials

Tech Talk

Forum home - Go back to Tech talk

 'SpeakUp' speech recognition module.
« Back · 1 · 2 · Next »
 Return to top of page · Post #: 1 · Written at 1:09:42 AM on 5 July 2014.
GTC's avatar
 GTC
 Location: Sydney, NSW
 Member since 28 January 2011
 Member #: 823
 Postcount: 6678

Couldn't help noticing the full page ad on the inside back cover of July issue of Silicon Chip for the SpeakUp speech recognition click board module, priced at US$39 plus shipping.

I already have far too many items on my projects list, but a few project ideas came to mind upon reading the specs for this nifty little device:
http://www.mikroe.com/click/speakup/.

I'm tempted while the dollar is still fairly strong. Will have to sleep on it.


 
 Return to top of page · Post #: 2 · Written at 9:08:10 AM on 5 July 2014.
Redxm's avatar
 Location: Tamworth, NSW
 Member since 6 April 2012
 Member #: 1126
 Postcount: 466

Oooh that looks like fun. I'm amazed at the price of 'modules' these days. Most of the time you couldn't buy the individual components for the same price.


 
 Return to top of page · Post #: 3 · Written at 9:18:27 AM on 5 July 2014.
Brad's avatar
 Administrator
 Location: Naremburn, NSW
 Member since 15 November 2005
 Member #: 1
 Postcount: 7290

I hope it works better than the robot at Pizza Hut. I was trying to order a pizza last week after spending two days shifting furniture I didn't feel like cooking. I was on the phone to Annie and to verify the address she asks what house number and suburb the caller lives in. For one reason or another she could recognise Ryde but not my three digit house number, regardless of whether I expressed it as a whole number or three separate digits it just wouldn't work.

I routinely use Bakelite PMG phones but for this exercise I had to plug in a tone dial phone and key in the house number. I suppose if we all had voices as clear as the sheila announcing train arrivals on all railway stations in NSW it wouldn't be an issue.


‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
A valve a day keeps the transistor away...

 
 Return to top of page · Post #: 4 · Written at 12:01:27 PM on 5 July 2014.
GTC's avatar
 GTC
 Location: Sydney, NSW
 Member since 28 January 2011
 Member #: 823
 Postcount: 6678

I hope it works better than the robot at Pizza Hut.

Actually, its name "SpeakUp" struck me as being amusing. I imagined the module replying "Speak Up!" when it can't understand the command.

I suppose if we all had voices as clear as the sheila announcing train arrivals on all railway stations in NSW it wouldn't be an issue.

... sure, when that system is used. I don't travel by train much these days but last time I did the station staff were doing announcements and there's no way I could understand what was being said due to a combination of accent, inability to articulate anyway, and poor amplification. I'd hate to have to rely on voice announcements.

Re use of bakelite phones, they didn't have the best microphones (or transmitters to use telco terminology) in them. Originally they came with granulated carbon types which were very poor performers especially once the carbon granules became compacted and started to "fry". Those were field-replaced by the PMG with more modern mics in the 1960s.

I'm thinking of going ahead and buying a SpeakUp board.


 
 Return to top of page · Post #: 5 · Written at 6:07:08 PM on 5 July 2014.
Redxm's avatar
 Location: Tamworth, NSW
 Member since 6 April 2012
 Member #: 1126
 Postcount: 466

Hows this go for an idea. Hook the speak up to your answering machine.
Then you could ring up and say 'turn the Kriesler on and boil the jug'


 
 Return to top of page · Post #: 6 · Written at 9:44:51 AM on 7 July 2014.
Maven's Gravatar
 Location: Canberra, ACT
 Member since 23 August 2012
 Member #: 1208
 Postcount: 584

.Redxm:Most of the time you couldn't buy the individual components for the same price.

About 90% of retail price now for small items is made up of stocking, cataloguing and handling costs along the chain from manufacturer to shopfront. Compare Jaycar prices to online vendors from Asia.

I've bought quite complex intelligent modules for less than the price of a few screws from a hardware store. We have to help pay the mortgages of every Australian with any role in the value chain.

Maven


 
 Return to top of page · Post #: 7 · Written at 5:26:42 PM on 10 July 2014.
Art's Gravatar
 Art
 Location: Somewhere, USA
 Member since 22 October 2013
 Member #: 1437
 Postcount: 896

I can't see the link, it says the page is unavailable.
I'm sceptical of anything that claims to understand a large vocabulary without specific voice training.
There have been some chips around since the 80's/90's that understood some vocab out of the box for all users, but only limited like "Stop, Left, Go, Forward, etc.".
I've played with free libraries Festival for speech synthesis and Sphinx for recognition of words from a dictionary,
and had both of those working on two different platforms, and they both still consume huge resource.


 
 Return to top of page · Post #: 8 · Written at 5:29:57 PM on 10 July 2014.
Brad's avatar
 Administrator
 Location: Naremburn, NSW
 Member since 15 November 2005
 Member #: 1
 Postcount: 7290

Try now. Smile


‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
A valve a day keeps the transistor away...

 
 Return to top of page · Post #: 9 · Written at 8:01:45 PM on 10 July 2014.
Art's Gravatar
 Art
 Location: Somewhere, USA
 Member since 22 October 2013
 Member #: 1437
 Postcount: 896

It more substantial than I expected Smile It's a good sign that it says "it learns from scratch to understand any dialect".
When using a freebie like Sphinx, unless you go through a complicated process to create a big data file, you can only select a number of words to look for out of a big dictionary by making an ASCII file consisting of only a limited number of trigger words.
It then turns out that an Australian has to fake a US accent to get the digits 5 and 9 correct.
The CPU is a generic microcontroller so someone must have got a lot of software componentry working together to do it well.


 
 Return to top of page · Post #: 10 · Written at 4:25:08 AM on 12 July 2014.
Wa2ise's avatar
 Location: Oradell, US
 Member since 2 April 2010
 Member #: 643
 Postcount: 830

QUOTE: It then turns out that an Australian has to fake a US accent to get the digits 5 and 9 correct.


Smile I didn't know we Americans had accents, except for Southerners or Boston. Smile And in New Jersey we have so little culture we don't have accents at all. Smile


 
 Return to top of page · Post #: 11 · Written at 9:45:26 AM on 12 July 2014.
Maven's Gravatar
 Location: Canberra, ACT
 Member since 23 August 2012
 Member #: 1208
 Postcount: 584

How's this: "Starrt worrk at fahv to nahn" v. "Staht werk at foive to noine"


 
 Return to top of page · Post #: 12 · Written at 7:10:14 PM on 12 July 2014.
Art's Gravatar
 Art
 Location: Somewhere, USA
 Member since 22 October 2013
 Member #: 1437
 Postcount: 896

Sphinx wants "N-AYE-N" and "F-AYE-VE"
https://www.youtube.com/watch?v=5XEHq4fHVPk.


 
 Return to top of page · Post #: 13 · Written at 8:26:38 PM on 12 July 2014.
Brad's avatar
 Administrator
 Location: Naremburn, NSW
 Member since 15 November 2005
 Member #: 1
 Postcount: 7290

It makes me wonder how the Croweaters would handle things with their slight Kiwi accents.


‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾‾
A valve a day keeps the transistor away...

 
 Return to top of page · Post #: 14 · Written at 6:49:15 PM on 13 July 2014.
Art's Gravatar
 Art
 Location: Somewhere, USA
 Member since 22 October 2013
 Member #: 1437
 Postcount: 896

If I hear good results from GTC, I might be close behind.. or maybe not even wait.
Just been playing with the Jaycar DTMF decoder chip to send codes for remote control over a radio.
For FM the signal might be clear enough and bandwidth good enough for speech to be recognised over the air.


 
 Return to top of page · Post #: 15 · Written at 7:26:55 PM on 13 July 2014.
GTC's avatar
 GTC
 Location: Sydney, NSW
 Member since 28 January 2011
 Member #: 823
 Postcount: 6678

According to DHL's tracking site, mine should be waiting at the post office on Monday.

Meanwhile I have confirmed from the manufacturer that the module can handle two or more voices issuing the same command, which can then be mapped to the given action. (A given action can affect each of the 12 GP IO ports.)

There have been some reports of users having difficulty with the command recording process when they used the auxiliary microphone jack. Manufacturer says to use directional mics as background noise can greatly affect the recording if omnidirectional mikes are used.


 
« Back · 1 · 2 · Next »
 You need to be a member to post comments on this forum.

Sign In

Username:
Password:
 Keep me logged in.
Do not tick box on a computer with public access.