Unscripted Speech Data Collection via Game-Based Interaction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for collecting natural speech data rely on scripted textual passages, which do not accurately represent real-world conversations, limiting the effectiveness of neural networks in understanding and processing actual speech dialogue.
Innovation Solution
A system and method that utilize a natural speech data generator to collect unscripted natural speech utterances through multi-player games, where users engage in conversational scenarios, capturing and analyzing their interactions to create a comprehensive dataset for training neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If scripted textual passages are used for data collection, then data collection efficiency is improved, but the naturalness and authenticity of speech data deteriorates
Solution Approach 1:
The system enables participants to generate natural speech data autonomously through their own spontaneous conversations during game play. The speech data is collected automatically without requiring external intervention or scripting, allowing participants to serve themselves in generating authentic speech samples while maintaining high collection efficiency
Solution Approach 2:
A game scenario acts as an intermediary mechanism that creates a natural conversational context. Instead of directly asking participants to read scripts or converse formally, the system mediates through game-based scenarios that naturally elicit spontaneous speech while maintaining structured data collection
2Device complexity
If scripted text is used for training neural networks, then training process simplicity is improved, but the effectiveness in processing real-world conversations deteriorates
Solution Approach 1:
The system changes the fundamental parameter of speech data authenticity by collecting unscripted, spontaneous speech instead of scripted text. This parameter change transforms the training data from artificial to authentic, enabling neural networks to learn from real conversational patterns while maintaining training process simplicity through automated collection
3Reliability
If unscripted natural speech is collected through games, then speech data authenticity is improved, but system complexity increases
Solution Approach 1:
The system achieves multi-functionality by combining game scenarios with speech data collection in a single integrated platform. The game engine serves multiple purposes: engaging participants, providing conversational context, and facilitating natural speech generation, thereby reducing overall system complexity despite the sophisticated data collection methodology
Data Source
AI summary
Example natural speech data generation systems and methods are described. In one implementation, a natural speech data generator initiates a game between a first player and a second player and determines a scenario associated with the game. A first role is assigned to the first player and a second role is assigned to the second player. The natural speech data generator receives multiple natural speech utterances by the first player and the second player during the game.