Phoneme aware method improves extreme speech quality restoration
P2Flow: Phoneme-aware Progressive Flow Matching for Extreme Speech Super-Resolution
Machine LearningSound
Summary
Audio recordings often lose sound quality when their frequencies are greatly reduced. The authors address this problem by creating a new method that uses speech sounds (phonemes) to help fill in missing details. Their approach also restores different frequency ranges step-by-step to rebuild clearer speech. After training, they refine the sound-producing model to make the output more natural. They tested their method on well-known speech datasets and found it performed better than previous techniques.
What this means in practice
- •For audio engineers: Enhance heavily degraded speech recordings by reconstructing lost frequencies using phoneme information and progressive restoration techniques.
- •For telecommunications teams: Improve voice quality in low-bandwidth or noisy communication channels by applying extreme speech super-resolution methods informed by phoneme data.
Authors
Ningyuan Yang, Yize Li, Pu Zhao, Diego A. Cuji, Kanad Sarkar, Ryan M. Corey, Xue Lin, Andrew C. Singer
Abstract
Generative models have recently demonstrated considerable promise in speech super-resolution (SSR). Nevertheless, the majority of existing work has concentrated on standard or versatile SSR configurations, leaving the extreme setting with severely limited spectral inputs largely unexplored. In this regime, current approaches exhibit marked performance degradation, underscoring the need for dedicated solutions. To bridge this gap, we introduce P2Flow, a phoneme-aware progressive flow matching (FM) framework designed for extreme SSR with three main strategies. First, our model leverages phonetic information to reconstruct missing spectral components. Furthermore, it employs a progressive architectural design that hierarchically restores distinct frequency regions. Finally, we incorporate post-training of the vocoder to enhance overall waveform fidelity. Extensive experiments are conducted on the TIMIT and VCTK datasets under both 1 kHz to 16 kHz and 2 kHz to 16 kHz settings, demonstrating that P2Flow yields state-of-the-art results across multiple evaluation metrics.