Signal Processing Apparatus for Conversation Detection in Noisy Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional signal processing methods struggle to accurately detect conversation establishment in informal daily conversations due to increased utterance overlap and the presence of laughter, leading to difficulties in distinguishing conversation partners from non-partners, especially in environments with background noise and self-talk.
Innovation Solution
A signal processing apparatus that separates mixed sound signals into individual excitations, performs speech detection, and calculates utterance overlap and silence durations, as well as laughter detection, to assess the degree of conversation establishment, thereby improving the detection of conversation partners in informal settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional voice/silence evaluation methods are used to detect conversation establishment, then the detection method is simple, but the detection accuracy deteriorates in informal conversations with utterance overlap and laughter
Solution Approach 1:
The patent segments the speech signal into multiple excitation sources using blind source separation (BSS) technology. This divides the mixed speech signal from multiple speakers into individual excitation signals, allowing separate analysis of each speaker's voice and silence patterns. The segmentation enables accurate detection of conversation establishment by evaluating voice/silence combinations for each excitation pair independently, resolving the contradiction between detection accuracy and processing complexity.
2Reliability
If temporal alternation of voice is used as the sole criterion for conversation detection, then the detection rule is simple, but the reliability deteriorates due to informal conversation patterns with overlap and laughter
Solution Approach 1:
The patent changes the evaluation parameters from simple temporal alternation to a comprehensive voice/silence evaluation system that considers multiple factors: voice presence, silence duration, and their temporal relationships. The system calculates a conversation establishment degree based on these parameter changes, allowing reliable distinction of conversation partners even in informal settings with utterance overlap and laughter, while adapting to various conversation patterns.
3Measurement precision
If blind source separation is applied to separate mixed speech signals, then the conversation partner detection accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent performs preliminary blind source separation to divide the mixed speech signal into excitation signals before conducting voice/silence evaluation. This preliminary action of separating the signals once allows subsequent evaluation to proceed efficiently on already-separated components, reducing repeated computational overhead and minimizing processing time loss while maintaining high separation precision.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A signal processing apparatus and signal processing method capable of correctly detecting that a conversation is established even in a daily environment are provided. In signal processing apparatus (100), excitation separation section (130) separates a mixed sound signal in which a plurality of excitations are mixed into the respective excitations. Speech detection section (140) performs speech detection on the plurality of separated excitation signals, judges whether or not the plurality of excitation signals are speech and generates speech section information indicating speech/non-speech information for each excitation signal. Identification parameter extraction section (150) extracts an identification parameter indicating a feature value of a daily conversation based on a plurality of excitation signals or speech section information. Conversation establishment degree calculation section (160) calculates and outputs a degree of establishment of a conversation based on the extracted identification parameter. Conversation partner identifying section (170) judges which excitation is a conversation partner using the degree of establishment of a conversation.