AMNet: An Acoustic Model Network for Enhanced Mandarin Speech Synthesis
Yubing Cao, Yinfeng Yu, Yongming Li, Liejun Wang
TL;DR
AMNet addresses Mandarin TTS challenges by integrating phrase structure awareness and local context modeling into a FastSpeech 2-based non-autoregressive framework. It introduces a phrase structure parser, a multi-kernel local convolution, and a tone-embedding module that decouples tone from phonemes, enabling explicit tone control. Empirical results show improvements in subjective naturalness (MOS) and objective measures such as a ~11.8% reduction in Mel-cepstral distortion and better F0 fitting captured by $F0(R^2)$. The work demonstrates the importance of local boundary information and tone-aware conditioning for high-quality Mandarin speech synthesis, suggesting strong potential for real-world TTS deployment.
Abstract
This paper presents AMNet, an Acoustic Model Network designed to improve the performance of Mandarin speech synthesis by incorporating phrase structure annotation and local convolution modules. AMNet builds upon the FastSpeech 2 architecture while addressing the challenge of local context modeling, which is crucial for capturing intricate speech features such as pauses, stress, and intonation. By embedding a phrase structure parser into the model and introducing a local convolution module, AMNet enhances the model's sensitivity to local information. Additionally, AMNet decouples tonal characteristics from phonemes, providing explicit guidance for tone modeling, which improves tone accuracy and pronunciation. Experimental results demonstrate that AMNet outperforms baseline models in subjective and objective evaluations. The proposed model achieves superior Mean Opinion Scores (MOS), lower Mel Cepstral Distortion (MCD), and improved fundamental frequency fitting $F0 (R^2)$, confirming its ability to generate high-quality, natural, and expressive Mandarin speech.