Browse Prior Art Database

Automatic Construction of a Syllabary for Infinite Vocabulary Speech Recognition

IP.com Disclosure Number: IPCOM000101059D
Original Publication Date: 1990-Jun-01
Included in the Prior Art Database: 2005-Mar-16
Document File: 2 page(s) / 79K

Publishing Venue

IBM

Related People

Bahl, LR: AUTHOR [+4]

Abstract

In small vocabulary speech recognition systems, it is customary to recognize speech one word at a time. For large or infinite vocabularies this word-based approach is inefficient and infeasible; too much computation and storage is required. A better approach is syllable-based recognition: the number of syllables required to cover a large vocabulary is substantially less than the number of words in that vocabulary.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Automatic Construction of a Syllabary for Infinite Vocabulary Speech Recognition

       In small vocabulary speech recognition systems, it is
customary to recognize speech one word at a time.  For large or
infinite vocabularies this word-based approach is inefficient and
infeasible; too much computation and storage is required.  A better
approach is syllable-based recognition:  the number of syllables
required to cover a large vocabulary is substantially less than the
number of words in that vocabulary.

      The definition of a syllable is arbitrary, but for speech
recognition purposes, the following properties are crucial:
1.   The decomposition of any given word into syllables is unique.
2.   The syllabary is as small as possible.

      The following invention provides a method of constructing a
syllabary suitable for speech recognition purposes.

      We will assume that a dictionary is available containing the
spellings and phonetic pronunciations of a large number of words:  at
least 50,000, preferably more. We will further assume that spelling/
phone alignments have been obtained as described in (1), and that
these alignments have been corrected as described in (2).

      The following definitions apply:
1.   The pronunciation of any given letter is the phone sequence
aligned against it.
2.   A letter tuple is a letter/pronunciation pair.
3.   A phone is a vowel if and only if it is normally expressed with
a stress level.
4.   A letter tuple is a vowel tuple if and only if there is a
vowel phone in the letter's pronunciation.
      The syllabary is constructed as follows.
Step 1.   Create a dictionary of inituples from the dictionary of
word spellings and pronunciations.
Step 2.   For each word in the dictionary of words perform Steps 3-9.
Step 3.   Place a syllable marker immediately following each vowel
tuple except the final one.
Step 4.   For each syllable marker in the curren...