# US 2019/0244064 A1 · Pattern recognition apparatus, method and medium

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

- Number: US 2019/0244064 A1
- Document ID: US20190244064A1
- Status: Published application
- Kind: A1
- Application number: 16/344,096
- Filed: 2016-10-25
- Published: 2019-08-08
- Earliest claimed: 2016-10-25
- Assignees: NEC Corporation
- Applicants: NEC Corporation
- Inventors: Shivangi MAHTO; Hitoshi YAMAMOTO; Takafumi KOSHINAKA
- Web page: https://patentgrep.com/patent/US20190244064A1

## CPC

- G06F 18/2411: based on the proximity to a decision surface, e.g. support vector machines
- G06F 18/2185: {the supervisor being an automated module, e.g. intelligent oracle}
- G06F 18/24: Classification techniques
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 20/00: Machine learning
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit
- G10L 2015/025: {Phonemes, fenemes or fenones being the recognition units}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

A pattern recognition apparatus according to the present invention includes: a memory; and at least one processor coupled to the memory, the processor performing operations, the operations comprising: transforming noisy feature vectors into denoised feature vectors; classifying the denoised feature vectors into related classes and estimated classes; calculating a cost by using the denoised feature vectors, clean feature vectors, the estimated classes, and feature vector labels; and updating parameters for transforming the noisy feature vectors according to the cost.

## Claims

### Claim 1

A pattern recognition apparatus comprising:
a memory; and
at least one processor coupled to the memory,
the processor performing operations, the operations comprising:
transforming noisy feature vectors into denoised feature vectors;
classifying the denoised feature vectors into related classes and estimated classes;
calculating a cost by using the denoised feature vectors, clean feature vectors, the estimated classes, and feature vector labels; and
updating parameters for transforming the noisy feature vectors according to the cost.

### Claim 2 (depends on 1)

The pattern recognition apparatus according to claim 1, wherein the operations further comprises
transforming testing feature vectors into the denoised feature vectors by using the updated parameters for transforming the noisy feature vector.

### Claim 3 (depends on 2)

The pattern recognition apparatus according to claim 2, wherein the operations further comprises
updating parameters for classifying the denoised feature vectors, and
classifying the denoised feature vectors by using the updated parameters for classifying the denoised feature vectors.

### Claim 4 (depends on 1)

The apparatus according to claim 1, wherein the operations further comprises
calculating the cost by using a transformation error between the denoised feature vectors and clean feature vectors and a classification error between estimated classes of noisy the noisy feature vectors and feature vector labels of classes.

### Claim 5 (depends on 1)

The apparatus according to claim 1, wherein the operations further comprises
transforming the noisy feature vectors by using a denoising autoencoder.

### Claim 6 (depends on 1)

The apparatus according to claim 1, wherein the operations further comprises
classifying the denoised feature vectors by using a support vector machine or a neural network.

### Claim 7 (depends on 4)

The apparatus according to claim 4, wherein
the transformation error is mean square error or cosine distance.

### Claim 8 (depends on 4)

The apparatus according to claim 4, wherein
the classification error is cross-entropy or margin error.

### Claim 9

A pattern recognition method comprising:
transforming noisy feature vectors into denoised feature vectors;
classifying the denoised feature vectors into related classes and estimated classes;
calculating a cost by using the denoised feature vectors, clean feature vectors, the estimated classes, and feature vector labels; and
updating parameters for transforming the noisy feature vector according to the cost.

### Claim 10

A non-transitory computer readable medium embodying a program, the program causing a pattern recognition apparatus to perform a method, the method comprising:
transforming noisy feature vectors into denoised feature vectors;
classifying the denoised feature vectors into related classes and estimated classes;
calculating a cost by using the denoised feature vectors, clean feature vectors, the estimated classes, and feature vector labels; and
updating parameters for transforming the noisy feature vector according to the cost.