Speech Recognition » Puneet Kalra

Puneet Kalra - www.puneetk.com - Socializing Robots

How To Improve Accuracy ( Sphinx 4 )

March 18th, 2010

Hello everyone,

I’m back with a very big question.

How to improve accuracy ?!?!?

Its been a week, I’m getting 2-3 mails everyday regarding “How to improve accuracy”. So this post is answer to all those and upcoming mails. First of all the problem is really big and complex, And to understand the problem first you need to understand the very basics of Speech Recognition and background process of Speech Recognition.

Basics of Speech Recognition are not Java coding part Or Whats gram file Or Why we use Recognizer and Microphone classes .. NO !

Remember ! Changes in Java Coding will not make any major effect in Accuracy. You need to focus on config.xml file.

To solve this problem, All you need to do is, understand the Automatic Speech Recognition, Search and Language Processing Algo’s. Specifically on Sphinx, you need understanding of Hidden Markov Models ( HMMs ) methods of Speech Recognition, Beam Searching, Language Configs and Modeling. Its highly recommended to read any good book before you continue to work on it.

Before we apply the configs on our application to give more accurate results, We must know to how to track results.
Click here to know how to apply Accuracy Tracker on your application.

Next part is, Knowing what kind of SR application you are creating. Basically, SR applications are divided into 2 major categories : Dictation and Command

Command Application When words are very few And Dictation Application When you have a large vocabulary.

And for both type of applications we need a language model or grammar file respectively. A language model is a file containing the probabilities of sequences of words. A grammar is a much smaller file containing sets of predefined combination of words.

Now again on the most important part, Accuracy Tracking. Find as many resources, examples, notes you can. Finally collect databases, Few are also available on Sphinx website. Work on it. Recognize it. Try to get your best results.

Expected Results in Silent Environment :
Command : 5% WER
Medium Vocabulary : 15% WER
Large Vocabulary : 30% WER

If environment is noisy, Multiply the results by 2.

If you are not getting the expected results. Keep trying and try to find the bugs in config file and run time reports.

Still if you are not happy with the results And have got extra time. Open the config.xml file and start playing with Configs and keep experimenting. Also, Make sure you make notes on every configs and their outputs.

If you have any better ideas or any suggestions, Please feel free to post comments.

That’s it for now ! Signing off.

5 Comments »

Posted in All, General

Tags: Accuracy Acoustic Models ASR Algo HMM Speech Recognition Sphinx 4.0

Basics of Java Speech Grammar Format ( JSGF )

March 12th, 2010

Hello Everyone,

This post is my response to those 4 help requests that i received in last few days. This one is going to be very basic and essentials of Java Speech Grammar Format ( JSGF ).

Every single file defines a single grammar only. Each grammar contains two parts:
The grammar header and the grammar body.

Grammar header format : #JSGF version char-encoding locale;

“#JSGF” is required and “version char-encoding locale;” is optional.

Grammar header example : #JSGF V1.0; & #JSGF V1.0 ISO8859-5 en;

After declaring Grammar Header, We need to specify the Grammar name.

Grammar name format : grammar grammarName;

This is a mandatory line. Else javax.speech.recognition.GrammarException will be thrown. Not only on Grammar Name, you will get GrammarExpection If you made any kind of mistake in grammar file.

Grammar Name Example : grammar helloWorld;

Once you are done with above part, you next step will be defining Grammar body.
The grammar body defines rules. you can define a rule only once. If you declare same rule twice then it will overwritten.

Rule Definition Format : public <ruleName> = ruleExpansion;

“public” is optional and remaining part is mandatory.
The rule expansion defines how the rule may be spoken. It is a logical combination of tokens (text that may be spoken) and references to other rules.

Rule Example : public <greet> = Hello;

The rule <greet> refers to a single token Hello. So to say rule <greet>, User must say word “Hello”.

A simple rule expansion can refer to one or more tokens or rules.

public <greet> = Hello;
public <completeGreet> = <greet> World;

Now, rule <completeGreet> refers to rule <greet> and token World. To say rule <completeGreet>, User must say “Hello World”.

Lets complete a simple “Hello World” grammar file.

#JSGF V1.0;

grammar simpleExample;

public <greet> = Hello;
public <completeGreet> = <greet> World;

This grammar file will allow you SR application to recognize 2 sentences. “Hello” and “Hello World”.

Now lets play a little bit with rule expansions.

Alternatives : “|”

public <greet> = Hello | Hey | Hi;

To say rule <greet>, User must say “Hello” or “Hey” or “Hi” But ONLY one of these three words.

Parentheses : “( )”

public <greet> = Hello ( World | User | Friend );
public <command> = ( Open | Close ) ( Door | Window );

To say rule <greet>, User must say “Hello World” or “Hello User” or “Hello Friend”.
And, to say rule <command>, User must say “Open Door” or “Open Window” or “Close Door” or “Close Window”.

Optional Grouping : “[ ]”

public <greet> = [ Hello ] World;

To say rule <greet>, User must say “Hello World” or “World”. As “Hello” is defined inside the Optional Grouping. So user may say it or not but “World” is mandatory.

Kleene Star : “*”

public <greet> = ( Hello | Hey | Hi )* World;

Any group or expansion followed by asterisk symbol indicates that it may be spoken zero or more times. For example “Hey Hello World” or “World” or “Hello Hello Hello World”.

Plus Operator : “+”

public <greet> = ( Hello | Hey | Hi )+ World;

A Plus Operator works same as “Kleene Star”, The only thing that makes difference is any group or expansion followed by plus symbol indicates that it may be spoken one ( NOT ZERO ) or more times.

You can also add comments in grammar file.

// One line comment
/* Multiple lines comment*/
/**
* Documentation comment
* @author Puneet Kalra
*/