Speech to Text

From BootstrapWiki
Jump to: navigation, search

Also known as Speach to Text...

February 2006

In answer to my question:

Can anyone recommend a speech to text solution? We're looking for an accuracy that is equivalent to taking notes, that is, not perfect.

I received the following advice:

I use the voice recognition built into Windows XP with very good results. Spend the money to get a good USB DSP headset. I bought the Planatronics xx-500 from Amazon for about $60 (it is about $130 retail) and have had very good accuracy. Plan to spend about 30 minutes training it for your voice. Integration into MS word is seamless.


There is a company that is call idictate.com -- they will transcribe for 1cent per word.

Here's an open source speech recognition package: http://cmusphinx.sourceforge.net/sphinx4/

And, since it's open source, you could easily put together something to automatically transcribe your podcasts, if you wanted to do that (and if you know Java).

There are other versions available at http://cmusphinx.sourceforge.net/html/cmusphinx.php.

Two of the best programs for a single user including multiple voices recording on the same PC:

1. Dragon Naturally Speaking

2. IBM Via Voice

They both take a bit of time to learn. Not only do you need to learn the program but it needs to learn to recognize your voice. Each time you correct a mis-interpreted word the system learns for the future.

They can learn to recognize multiple voices as well. They can translate real time or from a digital recording.

These do require a learning curve so if you are not using it a lot of times you may want to hire a dictation service.

I don't know what your budget is - but the best product on the market today is the IBM Via Voice which was acquired by Dragon Systems

Google Dragon Naturally Speaking

they have an entire array of products ranging from consumer to very elaborate industry specific fro medical, legal and much more and ranging in price from the 100 or so - to the several thousands - depending on industry, licensing, etc...

Take a look at this demo from Vista. If you’re in a hurry, skip to minute 5 or so:


It has been three years since I looked in this area. Two software I experimented with were IBM's ViaVoice and Scansoft's Dragon Naturally Dictate. Both claimed to have over 95% accuracy. I was on a Dragon Naturally Dictate user's list. From what I gathered, the quality of the microphone made a big difference. It is a matter of garbage in garbage out.

If you are looking for a software library to embed in an application checkout Microsoft's Speech SDK. Come to think of it, there may be a voice interface already built into Tablet PC OS.

Hope that helped.