# US 2019/0156853 A1 · Signal detection device, signal detection method, and signal detection program

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

- Number: US 2019/0156853 A1
- Document ID: US20190156853A1
- Status: Published application
- Kind: A1
- Application number: 15/759,571
- Filed: 2016-06-16
- Published: 2019-05-23
- Earliest claimed: 2015-09-16
- Assignees: NEC Corporation
- Applicants: NEC Corporation
- Inventors: Tatsuya KOMATSU; Yuzo SENDA
- Web page: https://patentgrep.com/patent/US20190156853A1

## CPC

- G10L 21/0308: characterised by the type of parameter measurement, e.g. correlation techniques, zero crossing techniques or predictive techniques
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F 18/2113: by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
- G06F 18/23213: with fixed number of clusters, e.g. K-means clustering
- G10L 25/51: for comparison or discrimination
- G10L 25/18: the extracted parameters being spectral information of each sub-band
- G10L 25/27: characterised by the analysis technique
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A signal detection device 10 includes a compression unit 11 which compresses an activation matrix by adding to each column an element of the row corresponding to a basis mapped to the information of the same acoustic element in an activation matrix computed by non-negative matrix factorization using a basis matrix, using the information of acoustic elements making up an acoustic event, mapped to the basis constituting the basis matrix.

## Claims

### Claim 1

A signal detection device comprising:
a compression unit configured to compress an activation matrix by adding to each column an element of the row corresponding to a basis mapped to the information of the same signal element in the activation matrix computed by non-negative matrix factorization using a basis matrix, using the information of signal elements making up a signal pattern, mapped to the basis constituting the basis matrix.

### Claim 2 (depends on 1)

The signal detection device according to claim 1, comprising:
a detection unit configured to detect a signal pattern included in a signal corresponding to a spectrogram formed of the activation matrix using the compressed activation matrix and a detection model used to detect the signal pattern.

### Claim 3 (depends on 1)

The signal detection device according to claim 1, comprising:
a generation unit configured to generate the basis matrix including the bases respectively corresponding to the signal elements by performing the non-negative matrix factorization to the spectrogram including the signal element forming the signal pattern so as to satisfy a predetermined condition.

### Claim 4 (depends on 3)

The signal detection device according to claim 3, comprising:
an analysis unit configured to perform the non-negative matrix factorization to a spectrogram associated with information indicating whether the corresponding signal is a signal to be detected using the basis matrix generated by the generation unit.

### Claim 5 (depends on 4)

The signal detection device according to claim 4, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.

### Claim 6

A signal detection method comprising:
compressing an activation matrix by adding to each column an element of the row corresponding to a basis mapped to the information of the same signal element in the activation matrix computed by non-negative matrix factorization using a basis matrix, using the information of signal elements making up a signal pattern, mapped to the basis constituting the basis matrix.

### Claim 7 (depends on 6)

The signal detection method according to claim 6, comprising:
detecting a signal pattern included in a signal corresponding to a spectrogram formed of the activation matrix using the compressed activation matrix and a detection model used to detect the signal pattern.

### Claim 8

A non-transitory computer-readable recording medium having recorded therein a signal detection program for causing a computer to perform compression processing of compressing an activation matrix by adding to each column an element of the row corresponding to a basis mapped to the information of the same signal element in the activation matrix computed by non-negative matrix factorization using a basis matrix, using the information of signal elements making up a signal pattern, mapped to the basis constituting the basis matrix.

### Claim 9 (depends on 8)

The signal detection program according to claim 8, for causing a computer to perform detection processing of detecting a signal pattern included in a signal corresponding to a spectrogram formed of the activation matrix using the compressed activation matrix and a detection model used to detect the signal pattern.

### Claim 10 (depends on 2)

The signal detection device according to claim 2, comprising:
a generation unit configured to generate the basis matrix including the bases respectively corresponding to the signal elements by performing the non-negative matrix factorization to the spectrogram including the signal element forming the signal pattern so as to satisfy a predetermined condition.

### Claim 11 (depends on 10)

The signal detection device according to claim 10, comprising:
an analysis unit configured to perform the non-negative matrix factorization to a spectrogram associated with information indicating whether the corresponding signal is a signal to be detected using the basis matrix generated by the generation unit.

### Claim 12 (depends on 2)

The signal detection device according to claim 2, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.

### Claim 13 (depends on 3)

The signal detection device according to claim 3, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.

### Claim 14 (depends on 4)

The signal detection device according to claim 4, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.

### Claim 15 (depends on 10)

The signal detection device according to claim 10, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.

### Claim 16 (depends on 11)

The signal detection device according to claim 11, comprising:
a learning unit configured to learn the detection model using the activation matrix including elements associated with information indicating the signal pattern included in the signal corresponding to the element.
