Gurudeva's concern on typing

[Guest guest]

It is the software from IBM.

http://www-4.ibm.com/software/speech/

 

Please have guruji(Sanjay) specify me that which one he would like to

have and I will have it mailed or send with someone.......

I am not sure how good are they but they are the most used in

business communities.

 

Om tat sat

 

Samir

[Guest guest]

Pranaam Sanjay and others,

 

I never used any commercial or free speech recognition software, but

I am familiar with the algorithms used and the state of technology.

 

Some speech recognition engines use neural networks approach and they

need to go through a learning period to become good at at. If you

spend time with the software and " train " it, it will start

recognizing your speech quite well.

 

Though this approach has following in the academic circles, the

industry prefers some other algorithms that are more computationally

intensive but do not need any learning period.

 

Products sold commercially to desktop users today (like the Dragon or

IBM products mentioned) are not that good, but I am familiar with

some state-of-the-art algorithms from SpeechWorks and Nuance that are

used in enterprise solutions. They're quite good. These engines have

a vocabulary of 20,000-30,000 words! With the processing power

available in the CPUs increasing everyday (333 MHz Pentium-2 was top

of the line a couple of years back. Now 2 GHz Pentium-4 has become

common), it is becoming possible to run more and more sophisticated

real-time audio, speech and video processing algorithms on computers.

 

At the new start-up company I joined in early May, we make mixed

network (standard phone and internet/IP) media servers with

integrated speech recognition, using which telecom carriers and

internet companies can build call centers and voice portals that have

speech recognition, speech-to-text, text-to-speech, interactive voice

response systems and other media services.

 

Speech recognition of this quality will be available for desktop

users very soon. For now, it may be wise to just type manually.

 

Narayan talked about speech recognition not being suitable for

Jyotish. Actually, some of the enterprise solutions come with

vocabulary files that are extendable. You can add additional words

and their phonetic representations in the files and then the speech

recognition engine will start recognizing those words too!

 

Just wait 2-3 years and Sanjay can indeed use speech recognition to

help him in typing mails!

 

BTW, those who are curious about speech recognition can try calling

in AT & T's TellMe portal at 1-800-555-TELL (see

http://www.tellme.com). You can get a lot of stock quotes, weather

reports, movie/restaurant information etc from their speech

recognition engine. It makes mistakes, but it's quite decent

(compared to what these things were 2 years back). This is based on

technology that is slightly inferior to state-of-the-art.

 

Your sishya,

Narasimha

 

> Impossible Solai!

> Dinanath made me go through that awesome speech recognition databank

> building once. I cannot go through it again. The computer starts

misbehaving

> and it acts as if some alien intelligence has got inside.

> Thanks for the concern.

> Best Wishes

> Sanjay Rath