# US 2014/0195232 A1 · Methods, systems, and circuits for text independent speaker recognition with automatic learning features

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

- Number: US 2014/0195232 A1
- Document ID: US20140195232A1
- Status: Published application
- Kind: A1
- Application number: 13/854,134
- Filed: 2013-04-01
- Published: 2014-07-10
- Earliest claimed: 2013-01-04
- Assignees: STMicroelectronics Asia Pacific Pte Ltd
- Inventors: Evelyn Kurniawati; Sapna George
- Web page: https://patentgrep.com/patent/US20140195232A1

## CPC

- G10L 17/04: Training, enrolment or model building
- G10L 15/063: {Training}
- G10L 15/144: {Training of HMMs}
- 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 17/06: Decision making techniques; Pattern matching strategies
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING

## Abstract

Embodiments provide a method and system of text independent speaker recognition with a complexity comparable to a text dependent version. The scheme exploits the fact that speech is a quasi-stationary signal and simplifies the recognition process based on this theory. The modeling allows the speaker profile to be updated progressively with the new speech sample that is acquired during usage time.

## Claims

### Claim 1

A method of text independent speaker recognition where the speaker profile is represented by clustered features extracted from user speech data during training session.

### Claim 2 (depends on 1)

The method of claim 1 where each of the clustered features is represented by a Gaussian mixture model, denoting a smaller unit of speech.

### Claim 3 (depends on 1)

The method of claim 1 where for each frame of incoming speech, log likelihood (loglk) is calculated for each cluster to find the probability that the speech is ending at that cluster.

### Claim 4 (depends on 3)

The method of claim 3 where the probability of speech ending at that cluster takes into account the probability of the cluster that precedes it and the transition probability to the current cluster.

### Claim 5 (depends on 4)

The method of claim 4 where transition probability between clusters are set lower than the transition probability to itself.

### Claim 6 (depends on 4)

The method of claim 4 where the transition probability between clusters is set at 0.05 and the transition probability to itself is set at 0.95.

### Claim 7 (depends on 1)

The method of claim 1 where the output decision can be taken at any point in time by evaluating the highest value of loglk variable among the speakers.

### Claim 8 (depends on 1)

The method of claim 1 where the output decision confidence value can be calculated by evaluating the difference between loglk variable among the two highest scoring speakers.

### Claim 9 (depends on 8)

The method of claim 8 where output confidence value can be the set as threshold to accept or reject a speaker, and the threshold value is based on the desired reliability of the system.

### Claim 10 (depends on 3)

The method of claim 3 where the value of loglk is reset to zero upon encountering a period of silence in the input speech.

### Claim 11 (depends on 3)

The method of claim 3 where the value of loglk is reset, upon encountering a period of silence in the input speech, to a value that corresponds to the speaker change probability according to a conversation state model.

### Claim 12 (depends on 1)

The method of claim 1 where GMM of the clustered features are continuously updated using speech frames that have high confidence value to belong to the corresponding speaker.