TTS Fluency Control for Spontaneous Speech Synthesis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional text-to-speech (TTS) systems generate speech that is non-spontaneous and lacks the diversity in fluency observed in human speech, failing to accurately simulate real-life conversations with pauses, repetitions, and varying speaking speeds.
Innovation Solution
Implementing a fluency control mechanism in TTS systems using a fluency reference factor to adjust acoustic features, incorporating fluency adjustment information and level representation to generate speech that mimics human spontaneity, including pauses and repetitions, tailored to different domains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional end-to-end TTS systems predict acoustic features based on text input, then speech waveform generation is achieved, but the synthesized speech lacks fluency diversity and spontaneity
Solution Approach 1:
The system dynamically adjusts fluency characteristics by introducing a fluency reference factor that modifies acoustic feature prediction based on desired fluency levels. This allows the TTS system to adapt between different speaking styles (spontaneous vs. non-spontaneous) while maintaining reliable speech generation, resolving the contradiction between fluency accuracy and fluency diversity
Solution Approach 2:
The invention changes the parameter space by adding a fluency reference factor as an additional control dimension. This parameter enables the system to generate speech with varying fluency characteristics (pauses, repetitions, speaking speed) without compromising the core speech synthesis functionality, thus achieving both reliability and adaptability
2Stability of the object's composition
If TTS systems generate speech with fixed fluency characteristics, then consistent speech output is achieved, but human-like spontaneity and natural variations are lost
Solution Approach 1:
The system transitions from fixed fluency characteristics to dynamic fluency control by incorporating a fluency reference factor. This enables consistent speech generation while allowing natural variations in fluency (pauses, repetitions, speed) that mimic human spontaneity, resolving the contradiction between stability and naturalness
3Ease of manufacture
If TTS systems add fluency control mechanisms, then spontaneous speech characteristics are improved, but system complexity increases
Solution Approach 1:
The fluency reference factor serves as an intermediary that bridges the text input and acoustic feature prediction. Instead of fundamentally redesigning the TTS system, this intermediary component modulates the existing acoustic prediction process to introduce fluency variations, achieving naturalness while minimizing additional complexity
Data Source
AI summary
The present disclosure provides methods and apparatuses for spontaneous text-to-speech (TTS) synthesis. A target text may be obtained. A fluency reference factor may be determined based at least on the target text. An acoustic feature corresponding to the target text may be generated with the fluency reference factor. A speech waveform corresponding to the target text may be generated based on the acoustic feature.