Speaker Identification via Clustering and Similarity Calculation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech processing technologies face challenges in accurately determining the number of speakers and identifying utterances from multiple speakers, as they require pre-classified utterances and struggle to differentiate between speakers effectively.

Innovation Solution

A speech processing device with an extractor, classifier, similarity calculator, and identifier is employed to extract speech features, classify utterances into clusters, calculate similarities with speaker models, and identify speakers based on weighted similarity sums, allowing for the identification of new speakers and integration of clusters.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If utterance data from multiple speakers is processed without pre-classification, then the system can handle mixed utterance data, but it becomes difficult to determine the number of speakers and identify utterances of the same speaker

Engineering Contradiction:

Improveability to handle mixed utterance dataVSAvoidspeaker identification accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The utterance data is segmented into multiple clusters based on speech features, where each cluster represents utterances from a single speaker. The clustering process divides the mixed utterance data into groups using algorithms that group similar speech patterns together, enabling automatic separation of different speakers without pre-classification

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

A speaker model serves as an intermediary between the raw utterance data and the identification process. The speaker model is learned from the clustered utterance data and acts as a reference template for identifying whether new utterances belong to known speakers or represent new speakers

Inventive Principle:

Principle #24Intermediary (Mediator)

2Adaptability or versatility

If speaker models are learned from clustered utterance data, then new speaker models can be identified, but the system complexity increases due to clustering and similarity calculation processes

Engineering Contradiction:

Improvenew speaker identification capabilityVSAvoidprocessing system complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary clustering of utterance data based on speech features before learning speaker models. This preliminary organization of data into speaker-specific clusters simplifies the subsequent speaker model learning process and enables efficient identification of new speakers by comparing against pre-established speaker models

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS10832685B2Speech processing device, speech processing method, and computer program product

Publication Date: 2020.11.10 KK TOSHIBA

AI summary

According to an embodiment, a speech processing device includes an extractor, a classifier, a similarity calculator, and an identifier. The extractor is configured to extract a speech feature from utterance data. The classifier is configured to classify the utterance data into a set of utterances for each speaker based on the extracted speech feature. The similarity calculator is configured to calculate a similarity between the speech feature of the utterance data included in the set and each of a plurality of speaker models. The identifier is configured to identify a speaker for each set based on the calculated similarity.