Speech Valence Detection via Emotion Word Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for detecting emotional valence from speech signals face limitations in classification performance, as they do not effectively differentiate between speech segments with and without emotion words, leading to inaccurate valence detection in real-world scenarios where emotions are interspersed with neutral segments.

Innovation Solution

A processor-implemented method and system that split input speech signals into segments based on the presence of emotion words, using a language-specific emotion words dictionary to select appropriate segments for training classifiers, thereby isolating emotion words-based and non-emotion words-based segments for training, which are then used to detect emotional valence in input speech signals.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If speech signals are split into many shorter segments for training, then the quantity of training data increases, but the measurement precision of emotional valence detection deteriorates because segments may not carry the same emotion as the overall story

Engineering Contradiction:
Improvequantity of training segmentsVSAvoidaccuracy of emotional valence detection
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The patent applies segmentation by dividing speech signals into segments based on the presence of emotion words. Instead of uniform splitting, segments are created dynamically around emotion words using a window approach (e.g., ±1 second), ensuring each segment captures complete emotional expressions while maintaining appropriate quantity for training.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by treating segments containing emotion words differently from other segments. Emotion word-based segments are extracted and handled separately with different training strategies, allowing the system to focus computational resources on segments with higher emotional information density while maintaining overall detection accuracy.

Inventive Principle:
Principle #3Local quality

2Productivity

If all speech segments are used for training classifiers, then the productivity of training data utilization increases, but the reliability of emotional valence detection deteriorates due to inclusion of neutral segments that do not carry emotional information

Engineering Contradiction:
Improvetraining data utilization efficiencyVSAvoidaccuracy of emotional valence detection
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent extracts segments containing emotion words from the overall speech signal using a language-specific emotion words dictionary. This extraction process isolates emotionally informative segments for targeted training, removing neutral segments that would otherwise dilute the training quality while maintaining high productivity through efficient segment selection.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter of segment selection from uniform random sampling to emotion-word-based selective sampling. By using emotion words as selection criteria and adjusting segment window parameters (e.g., ±1 second around emotion words), the system optimizes both training productivity and detection reliability.

Inventive Principle:
Principle #35Parameter changes

3Device complexity

If emotion words are not differentiated from other words, then the device complexity is reduced, but the measurement precision of emotional valence detection deteriorates

Engineering Contradiction:
Improvecomplexity of speech processing systemVSAvoidaccuracy of emotional valence detection
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent introduces an intermediary component: a language-specific emotion words dictionary. This dictionary acts as a mediator between the raw speech signal and the classification process, enabling precise identification of emotionally informative segments without requiring complex deep learning architectures, thus maintaining relatively simple device complexity while improving measurement precision.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3989224B1Selection of speech segments for training classifiers for detecting emotional valence from input speech signals
Publication Date: 2024.07.31 TATA CONSULTANCY SERVICES LTD
  • EP3989224B1 patent drawingFigure 1
  • EP3989224B1 patent drawingFigure 2
  • EP3989224B1 patent drawingFigure 3

AI summary

An important task in several wellness applications is detection of emotional valence from speech. Two types of features of speech signals are used to detect valence: acoustic features and text features. Acoustic features are derived from short frames of speech, while text features are derived from the text transcription. Present disclosure provides systems and methods that determine the effect of text on acoustic features. Acoustic features of speech segments carrying emotion words are to be treated differently from other segments that do not carry such words. Only specific speech segments of the input speech signal are considered based on a dictionary specific to a language to assess emotional valence. A model trained (or trained classifier) for specific language either by including the acoustic features of the emotion related words or by omitting it is used by the system for determining emotional valence in an input speech signal.