Generation device, recognition system, and generation method for generating finite state transducer

• US 10,600,407 B2
• Filed: 02/09/2017
• Issued: 03/24/2020
• Est. Priority Date: 07/20/2016
• Status: Active Grant

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Nagao, Manabu
• Primary Examiner(s): Zhu, Richard Z
• Application Number:
  US15/428,648
• Publication Number:
  US 20180025723A1
• Time in Patent Office: 1,139 Days
• Field of Search: None
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/14 using statistical models, e.g. Hidden Markov Models [HMMs] G10L15/18 takes precedence
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/148 Duration modelling in HMMs, e.g. semi HMM, segmental models or transition probabilities
• CPC Class Codes: G10L 15/1807 using prosody or stress
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/027 Syllables being the recognition units

Abstract

A generation device includes a receiving unit and a generating unit. The receiving unit receives a model representing correspondence between one or more phonetic symbols and one or more words. The generating unit generates a first finite state transducer based on the model, the first finite state transducer at least including, as outgoing transitions from a first state representing transition destination of a first transition which has a first phonetic symbol of a predetermined type as input symbol, a second transition that has a second phonetic symbol, which is different than a particular symbol representing part or whole of input symbol of the first transition, as input symbol, and a third transition that has a third phonetic symbol, which represents the particular symbol or silence, as input symbol.

Claims

1. A recognition system comprising a generation device and a recognition device wherein:

the generation device comprises;

a receiving unit that receives a model representing correspondence between one or more phonetic symbols and one or more words; and

a generating unit that generates a first finite state transducer corresponding to slow utterances based on the model, the first finite state transducer at least including, as outgoing transitions from a first state, a second transition, and a third transition,

the first state representing transition destination of a first transition which has a first phonetic symbol of a predetermined type as input symbol, wherein the predetermined type comprises a slowly utterable vowel or syllabic nasal,

the second transition that has a second phonetic symbol, which is different than a particular symbol representing part or whole of input symbol of the first transition, as input symbol, and

the third transition that has a third phonetic symbol, representing the particular symbol or silence, as input symbol, and the recognition device comprises;

a searching unit that refers to the first finite state transducer and recognizes a word corresponding to an input speech.

2. The recognition system according to claim 1, wherein the phonetic symbol is a phoneme,

the first phonetic symbol is a particular phoneme of the predetermined type, and

the particular symbol represents pronunciation of the particular phoneme that is the input symbol of the first transition.

3. The recognition system according to claim 1, wherein

the phonetic symbol is a syllable including one or more phonemes,

the first phonetic symbol is a syllable in which a particular phoneme of the predetermined type is included at end, and

the particular symbol represents pronunciation of last

of the particular phoneme included in the input symbol of the first transition.

4. The recognition system according to claim 1, wherein transition destination of the third transition is the first state.

5. The recognition system according to claim 4, wherein the generating unit generates the first finite state transducer in which a weight greater than zero is assigned to the third transition.

6. The recognition system according to claim 1, wherein the generating unit generates the first finite state transducer that includes, as outgoing transitions from a second state representing transition destination of the third transition,

a fourth transition, which has either the first phonetic symbol or the third phonetic symbol as input and which represents transition to the second state, and

a fifth transition, which has the second phonetic symbol as input.

7. The recognition system according to claim 6, wherein the generating unit generates the first finite state transducer in which a weight greater than zero is assigned to the fourth transition.

8. The recognition system according to claim 1, wherein the first phonetic symbol comprises a slowly utterable vowel.

9. The recognition system according to claim 1, further comprising a composing unit that composes

the first finite state transducer,

a second finite state transducer generated based on a language model, and

a third finite state transducer generated based on an acoustic model to generate a fourth finite state transducer.

10. The recognition system according to claim 1, wherein the model is

a lexicon, which represents correspondence between one or more phonetic symbols and one or more words, or

a finite state transducer, which converts one or more phonetic symbols into words.

11. A generation method for generating finite state transducer, comprising:

receiving a model representing correspondence between one or more phonetic symbols and one or more words;

generating a first finite state transducer corresponding to slow utterances based on the model, the first finite state transducer at least including, as outgoing transitions from a first state, a second transition, and a third transition,

the first state representing transition destination of a first transition which has a first phonetic symbol of a predetermined type as input symbol, wherein the predetermined type comprises a slowly utterable vowel or syllabic nasal,

the second transition that has a second phonetic symbol, which is different than a particular symbol representing part or whole of input symbol of the first transition, as input symbol, and

the third transition that has a third phonetic symbol, representing the particular symbol or silence, as input symbol, and

referring to the first finite state transducer and recognizes a word corresponding to an input speech.