# US 10,319,373 B2 · Information processing device, information processing method, computer program product, and recognition system

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 10,319,373 B2
- Document ID: US10319373B2
- Status: Granted
- Kind: B2
- Application number: 15/390,232
- Filed: 2016-12-23
- Published: 2017-09-14
- Granted: 2019-06-11
- Earliest claimed: 2016-03-14
- Assignees: Kabushiki Kaisha Toshiba
- Applicants: Kabushiki Kaisha Toshiba
- Inventors: Manabu Nagao
- Web page: https://patentgrep.com/patent/US10319373B2

## CPC

- G10L 15/142: {Hidden Markov Models [HMMs]}
- G10L 15/144: {Training of HMMs}
- G10L 15/187: Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

An information processing device includes a phonetic converting unit, an HMM converting unit, and a searching unit. The phonetic converting unit converts a phonetic symbol sequence into a hidden Markov model (HMM) state sequence in which states of an HMM are aligned. The HMM converting unit converts the HMM state sequence into a score vector sequence indicating the degree of similarity to a specific pronunciation using a similarity matrix defining the similarity between the states of the HMM. The searching unit searches for a path having a better score for the score vector sequence than that of the other paths out of paths included in a search network and outputs a phonetic symbol sequence corresponding to the retrieved path.

## Claims

### Claim 1

An information processing device comprising:
one or more processors electrically coupled to a memory and configured to:
convert a phonetic symbol sequence into a hidden Markov model (HMM) state sequence in which states of the HMM are aligned;
convert the HMM state sequence into a score vector sequence indicating a degree of similarity to a specific pronunciation using a similarity matrix defining similarity between the states of the HMM; and
search for a path having a better score for the score vector sequence than a score of other paths out of paths included in a search network and output a phonetic symbol sequence corresponding to the path that is searched for, wherein:
the search network is a weighted finite state transducer generated using a first weighted finite state transducer having a restriction on the phonetic symbol sequence to be output, and
the first weighted finite state transducer receives, when N (N is an integer of 1 or larger) is the number of types of phonetic symbol, the phonetic symbol sequence alone in which the number of types of the phonetic symbol sequence having a length of M (M is an integer of 1 or larger) is smaller than NM.

### Claim 2 (depends on 1)

The information processing device according to claim 1, wherein the search network is a weighted finite state transducer obtained by composing the first weighted finite state transducer, a second weighted finite state transducer indicating the HMM, and a third weighted finite state transducer that converts a context-dependent phonetic symbol into a context-independent phonetic symbol.

### Claim 3 (depends on 1)

The information processing device according to claim 1, wherein the one or more processors are further configured to search for the path while considering that a value of an i-th dimension (i is an integer of 1 or larger) of a score vector included in the score vector sequence is an output probability of a state i in the HMM.

### Claim 4

A recognition system comprising an information processing device and a recognition device, wherein:
the information processing device comprising:
one or more first processors electrically coupled to a first memory and configured to:
convert a phonetic symbol sequence of an added word into an HMM state sequence in which states of the HMM are aligned;
convert the HMM state sequence into a score vector sequence indicating a degree of similarity to a specific pronunciation using a similarity matrix defining similarity between the states of the HMM; and
search for a path having a better score for the score vector sequence than a score of other paths out of paths included in a search network and output a phonetic symbol sequence corresponding to the path that is searched for, wherein:
the search network is a weighted finite state transducer generated using a first weighted finite state transducer having a restriction on the phonetic symbol sequence to be output, and
the first weighted finite state transducer receives, when N (N is an integer of 1 or larger) is the number of types of phonetic symbol, the phonetic symbol sequence alone in which the number of types of the phonetic symbol sequence having a length of M (M is an integer of 1 or larger) is smaller than NM, and
the recognition device comprising:
one or more second processors electrically coupled to a second memory and configured to:
set the phonetic symbol sequence output by the one or more first processors as a phonetic symbol sequence of the added word; and
register the added word as a word to be recognized.

### Claim 5

An information processing method comprising:
converting a phonetic symbol sequence into an HMM state sequence in which states of the HMM are aligned;
converting the HMM state sequence into a score vector sequence indicating a degree of similarity to a specific pronunciation using a similarity matrix defining similarity between the states of the HMM; and
searching for a path having a better score for the score vector sequence than a score of other paths out of paths included in a search network and outputting a phonetic symbol sequence corresponding to the path that is searched for, wherein:
the search network is a weighted finite state transducer generated using a first weighted finite state transducer having a restriction on the phonetic symbol sequence to be output, and
the first weighted finite state transducer receives, when N (N is an integer of 1 or larger) is the number of types of phonetic symbol, the phonetic symbol sequence alone in which the number of types of the phonetic symbol sequence having a length of M (M is an integer of 1 or larger) is smaller than NM.