# US 2017/0287489 A1 · Synthetic oversampling to enhance speaker identification or verification

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

- Number: US 2017/0287489 A1
- Document ID: US20170287489A1
- Status: Published application
- Kind: A1
- Application number: 15/088,500
- Filed: 2016-04-01
- Published: 2017-10-05
- Earliest claimed: 2016-04-01
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Narayan Biswal; Gokcen Cilingir; Barnan Das
- Web page: https://patentgrep.com/patent/US20170287489A1

## CPC

- G10L 17/06: Decision making techniques; Pattern matching strategies
- G10L 17/02: Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
- G10L 17/04: Training, enrolment or model building
- G10L 17/08: Use of distortion metrics or a particular distance between probe pattern and reference templates
- G10L 17/24: the user being prompted to utter a password or a predefined phrase
- G10L 25/78: Detection of presence or absence of voice signals (switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

An apparatus for oversampling audio signals is described herein. The apparatus includes one or more microphones to receive audio signals and an extractor to extract a set of feature points from the audio signals. The apparatus also includes a processing unit to determine a distance between each pair of feature points and an oversampling unit to generate a plurality of new feature points based the distance between each pair of feature points.

## Claims

### Claim 1

An apparatus, for oversampling audio signals comprising:
one or more microphones to receive audio signals;
an extractor to extract a set of feature points from the audio signals;
a processing unit to determine a distance between each pair of feature points of the set of feature points; and
an oversampling unit to generate a plurality of new feature points based on the distance between each pair of feature points and to generate a speaker voice model.

### Claim 2 (depends on 1)

The apparatus of claim 1, wherein a plurality of feature vectors are created using the set of feature points and the plurality of new feature points.

### Claim 3 (depends on 2)

The apparatus of claim 2, wherein the plurality of feature vectors are determined to be valid according to voice activity detection.

### Claim 4 (depends on 1)

The apparatus of claim 1, wherein the generation of the plurality of new feature points generates new feature points that number at least 50% of a number of feature points in the set of feature points.

### Claim 5 (depends on 1)

The apparatus of claim 1, wherein the distance between each pair of feature points is a Euclidean distance.

### Claim 6 (depends on 1)

The apparatus of claim 1, wherein the oversampling unit is to generate the plurality of new feature points based the distance between each pair of feature points, wherein each pair of feature points are valid feature points according to voice activity detection.

### Claim 7 (depends on 1)

The apparatus of claim 1, wherein the oversampling is a pre-processing function of a voice biometric system.

### Claim 8 (depends on 1)

The apparatus of claim 1, wherein the speaker voice model is based on the set of feature points and the plurality of new feature points.

### Claim 9 (depends on 1)

The apparatus of claim 1, wherein an authentication decision is based on the set of feature points and the plurality of new feature points.

### Claim 10 (depends on 1)

The apparatus of claim 1, wherein the oversampling results in low false rejection and false acceptance rates.

### Claim 11 (depends on 1)

The apparatus of claim 1, wherein an utterance length for audio signal capture is directly proportional to the set of feature points plus the plurality of new feature points.

### Claim 12

An method for oversampling audio signals, comprising:
capturing audio signals from an utterance;
extracting a set of feature points from the audio signals;
determining a distance between each pair of feature points of the set of feature points;
generating a plurality of new feature points based on the distance between each pair of feature points of the set of feature points; and
identifying a user via recognition decisions based on the plurality of new feature points and the set of feature points.

### Claim 13 (depends on 12)

The method of claim 12, wherein the utterance is a phrase independent utterance.

### Claim 14 (depends on 12)

The method of claim 12, wherein the utterance is a phrase dependent utterance.

### Claim 15 (depends on 12)

The method of claim 12, wherein the set of feature points and the plurality of new feature points are used to train a speaker model.

### Claim 16

(canceled)

### Claim 17 (depends on 12)

The method of claim 12, wherein an utterance length for audio signal capture is directly proportional to the set of feature points plus the plurality of new feature points.

### Claim 18 (depends on 12)

The method of claim 12, wherein the plurality of new feature points and the set of feature points are Mel-frequency cepstral coefficients (MFCC).

### Claim 19

A system for oversampling audio signals comprising, comprising:
a microphone;
a memory that is to store instructions and that is communicatively coupled to the microphone; and
a processor communicatively coupled to the microphone and the memory, wherein when the processor is to execute the instructions, the processor is to:
receive audio signals from an utterance;
extract a set of feature points from the audio signals;
determine a distance between each pair of feature points of the set of feature points
generating a plurality of new feature points based on the distance between each pair of feature points of the set of feature points;
extrapolate a plurality of feature vectors based on the plurality of new feature points and the set of feature points; and
identify a user via recognition decisions based on the plurality of new feature points, the set of feature points, and the plurality of feature vectors.

### Claim 20 (depends on 19)

The system of claim 19, wherein the processor is to build a speaker voice model based on the plurality of new feature points and the set of feature points.

### Claim 21 (depends on 19)

The system of claim 19, wherein the plurality of new feature points and the set of feature points are Mel-frequency cepstral coefficients (MFCC), linear predictive cepstral coefficients (LPCCs), line spectral frequencies (LSFs), and perceptual linear prediction (PLP) coefficients, or any combination thereof.

### Claim 22 (depends on 19)

The system of claim 19, wherein the plurality of feature vectors are determined to be valid according to voice activity detection.

### Claim 23

A non-transitory computer-readable medium, comprising instructions that, when executed by the processor, direct the processor to
capture audio signals from an utterance;
extract a set of feature points from the audio signals;
determine a distance between each pair of feature points of the set of feature points;
generate a plurality of new feature points based on the distance between each pair of feature points of the set of feature points; and
identify a user via recognition decisions based on the plurality of new feature points and the set of feature points.

### Claim 24 (depends on 23)

The non-transitory computer-readable medium of claim 23, wherein the utterance is a phrase independent utterance.

### Claim 25 (depends on 23)

The non-transitory computer-readable medium of claim 23, wherein the utterance is a phrase dependent utterance.
