LetterToSoundImpl (FreeTTS 1.2)

Overview

Package

Class

Tree

Deprecated

Index

Help

PREV CLASS NEXT CLASS

FRAMES NO FRAMES

SUMMARY: NESTED | FIELD | CONSTR | METHOD

DETAIL: FIELD | CONSTR | METHOD

com.sun.speech.freetts.lexicon
Class LetterToSoundImpl

java.lang.Object
  com.sun.speech.freetts.lexicon.LetterToSoundImpl

All Implemented Interfaces:: LetterToSound

public class LetterToSoundImpl
extends java.lang.Object
implements LetterToSound

Provides the phone list for words using the CMU6 letter-to-sound (LTS) rules, which are based on the Black, Lenzo, and Pagel paper, "Issues in Building General Letter-to-Sound Rules." Proceedings of ECSA Workshop on Speech Synthesis, pages 77-80, Australia, 1998.

The LTS rules are a simple state machine, with one entry point for each letter of the alphabet (lower case letters are always assumed, and the rules keep an array with one entry per letter that point into the state machine).

The state machine consists of a huge array, with most entries containing a decision and the indeces of two other entries. The first of these two indeces represents where to go if the decision is true, and the second represents where to go if the decision is false. All entries that do not contain a decision are final entries, and these contain a phone.

The decision in this case is a simple character comparison, but it is done in the context of a window around the character in the word. The decision consists of a index into the context window and a character value. If the character in the context window matches the character value, then the decision is true.

The machine traversal for each letter starts at that letter's entry in the state machine and ends only when it reaches a final state. If there is no phone that can be mapped, the phone in the final state is set to 'epsilon.'

The context window for a character is generated in the following way:

Pad the original word on either side with '#' and '0' characters the size of the window for the LTS rules (in this case, the window size is 4). The "#" is used to indicate the beginning and end of the word. So, the word "monkey" would turn into "000#monkey#000".
For each character in the word, the context window consists of the characters in the padded form the preceed and follow the word. The number of characters on each side is dependent upon the window size. So, for this implementation, the context window for the 'k' in monkey is "#money#0".

Here's how the phone for 'k' in 'monkey' might be determined:

Create the context window "#money#0".
Start at the state machine entry for 'k' in the state machine.
Grab the 'index' from the current state. This represents an index into the context window.
Compare the value of the character at the index in the context window to the character from the current state. If there is a match, the next state is the qtrue value. If there isn't a match, the next state is the qfalse state.
Keep on working through the machine until you read a final state.
When you get to the final state, the phone is the character in that state.

This implementation will either read from a straight ASCII file or a binary file. When reading from an ASCII file, you can specify when the input line is tokenized: load, lookup, or never. If you specify 'load', the entire file will be parsed when it is loaded. If you specify 'lookup', the file will be loaded, but the parsing for each line will be delayed until it is referenced and the parsed form will be saved away. If you specify 'never', the lines will parsed each time they are referenced. The default is 'load'. To specify the load type, set the system property as follows:

   -Dcom.sun.speech.freetts.lexicon.LTSTokenize=load

[[[TODO: This implementation uses ASCII 'a'-'z', which is not internationalized.]]]

Field Summary
`protected java.util.HashMap`	`letterIndex` The indexes of the starting points for letters in the state machine.
`protected boolean`	`tokenizeOnLoad` If true, the state string is tokenized when it is first read.
`protected boolean`	`tokenizeOnLookup` If true, the state string is tokenized the first time it is referenced.

Constructor Summary
`LetterToSoundImpl(java.net.URL ltsRules, boolean binary)` Class constructor.

Method Summary
`boolean`	`compare(LetterToSoundImpl other)` Compares this LTS to another for debugging purposes.
`void`	`dumpBinary(java.lang.String path)` Dumps a binary form of the letter to sound rules.
`protected char[]`	`getFullBuff(java.lang.String word)` Makes a character array that looks like "000#word#000".
`java.lang.String[]`	`getPhones(java.lang.String word, java.lang.String partOfSpeech)` Calculates the phone list for a given word.
`protected com.sun.speech.freetts.lexicon.LetterToSoundImpl.State`	`getState(int i)` Gets the `State` at the given index.
`protected com.sun.speech.freetts.lexicon.LetterToSoundImpl.State`	`getState(java.lang.String s)` Gets the `State` based upon the `String`.
`protected com.sun.speech.freetts.lexicon.LetterToSoundImpl.State`	`getState(java.lang.String type, java.util.StringTokenizer tokenizer)` Gets the `State` based upon the `type` and `tokenizer.`
`static void`	`main(java.lang.String[] args)` Translates between text and binary forms of the CMU6 LTS rules.
`protected void`	`parseAndAdd(java.lang.String line)` Creates a word from the given input line and add it to the state machine.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

tokenizeOnLoad

protected boolean tokenizeOnLoad

If true, the state string is tokenized when it is first read. The side effects of this are quicker lookups, but more memory usage and a longer startup time.

tokenizeOnLookup

protected boolean tokenizeOnLookup

If true, the state string is tokenized the first time it is referenced. The side effects of this are quicker lookups, but more memory usage.

letterIndex

protected java.util.HashMap letterIndex

The indexes of the starting points for letters in the state machine.

Constructor Detail