Arabic TTS with FastPitch: Reproducible Baselines, Adversarial Training, and Oversmoothing Analysis

TL;DR

This paper develops reproducible FastPitch-based Arabic TTS baselines, introduces new metrics for oversmoothing analysis, and employs adversarial training and synthetic data augmentation to improve prosody and stability.

Contribution

It presents the first reproducible FastPitch Arabic TTS baseline, introduces cepstral-domain oversmoothing metrics, and enhances multi-speaker TTS with adversarial loss and synthetic voice augmentation.

Findings

Adversarial spectrogram loss reduces oversmoothing effectively.

Cepstral metrics reveal oversmoothing effects during training.

Synthetic voices improve prosodic diversity in multi-speaker TTS.

Abstract

Arabic text-to-speech (TTS) remains challenging due to limited resources and complex phonological patterns. We present reproducible baselines for Arabic TTS built on the FastPitch architecture and introduce cepstral-domain metrics for analyzing oversmoothing in mel-spectrogram prediction. While traditional Lp reconstruction losses yield smooth but over-averaged outputs, the proposed metrics reveal their temporal and spectral effects throughout training. To address this, we incorporate a lightweight adversarial spectrogram loss, which trains stably and substantially reduces oversmoothing. We further explore multi-speaker Arabic TTS by augmenting FastPitch with synthetic voices generated using XTTSv2, resulting in improved prosodic diversity without loss of stability. The code, pretrained models, and training recipes are publicly available at: https://github.com/nipponjo/tts-arabic-pytorch.