mlsanityData to 19 Oct 2026

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

arXiv 2512.00937 ↗First posted 30 Nov 2025eess.AScited by 1

On the map: Speech and audio language models → Multilingual and low-resource ASR

Abstract (arXiv)

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.

Key facts. Cited by 1 paper in the map (data through 19 Oct 2026). Placed in Multilingual and low-resource ASR, one of 11 directions in Speech and audio language models. Builds on 13 of its 28 references that are in the map. First posted on arXiv 30 Nov 2025.

Builds on

13 of 28 references.

References · 13

  1. 2018Spectral Normalization for Generative Adversarial Networkscited by 1,666
  2. 2020HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesiscited by 1,462
  3. 2016Least Squares Generative Adversarial Networkscited by 1,044
  4. 2017Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictionscited by 1,023
  5. 2021Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speechcited by 737
  6. 2018Free-Form Image Inpainting with Gated Convolutioncited by 597
  7. 2024XTTS: a Massively Multilingual Zero-Shot Text-to-Speech Modelcited by 239
  8. 2023Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesiscited by 218
  9. 2020FastPitch: Parallel Text-to-speech with Pitch Predictioncited by 204
  10. 2021One TTS Alignment To Rule Them Allcited by 63
  11. 2018Fast Spectrogram Inversion using Multi-head Convolutional Neural Networkscited by 28
  12. 2019Neural Arabic Text Diacritization: State of the Art Results and a Novel Approach for Machine Translationcited by 12
1 more
  1. 2024CATT: Character-based Arabic Tashkeel Transformercited by 5

Built on by

Cited by year (2026 to 19 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026An Ultra-Low Latency, End-to-End Streaming Speech Synthesis Architecture via Block-Wise Generation and Depth-Wise Codec Decodingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026More Data, Fewer Diacritics: Scaling Arabic TTS
  2. 2023ClArTTS: An Open-Source Classical Arabic Text-to-Speech Corpus
  3. 2023Unsupervised Data Selection for TTS: Using Arabic Broadcast News as a Case Study
  4. 2026Habibi: Laying the Open-Source Foundation of Unified-Dialectal Arabic Speech Synthesis
  5. 2026LLM-to-Speech: A Synthetic Data Pipeline for Training Dialectal Text-to-Speech Models
  6. 2025ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis
  7. 2024Towards Zero-Shot Text-To-Speech for Arabic Dialects
  8. 2022SANE-TTS: Stable And Natural End-to-End Multilingual Text-to-Speech