mlsanityData to 14 Oct 2026

Regotron: Regularizing the Tacotron2 architecture via monotonic alignment loss

arXiv 2204.13437 ↗First posted 28 Apr 2022cs.SD cs.LG eess.AScited by 3

Not on the map. Why?

Abstract (arXiv)

Recent deep learning Text-to-Speech (TTS) systems have achieved impressive performance by generating speech close to human parity. However, they suffer from training stability issues as well as incorrect alignment of the intermediate acoustic representation with the input text sequence. In this work, we introduce Regotron, a regularized version of Tacotron2 which aims to alleviate the training issues and at the same time produce monotonic alignments. Our method augments the vanilla Tacotron2 objective function with an additional term, which penalizes non-monotonic alignments in the location-sensitive attention mechanism. By properly adjusting this regularization term we show that the loss curves become smoother, and at the same time Regotron consistently produces monotonic alignments in unseen examples even at an early stage (13% of the total number of epochs) of its training process, whereas the fully converged Tacotron2 fails to do so. Moreover, our proposed regularization method has no additional computational overhead, while reducing common TTS mistakes and achieving slighlty improved speech naturalness according to subjective mean opinion scores (MOS) collected from 50 evaluators.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 23 of its 32 references that are in the map. First posted on arXiv 28 Apr 2022.

Builds on

23 of 32 references.

References · 23

  1. 2017Attention Is All You Needcited by 60,361
  2. 2014Sequence to Sequence Learning with Neural Networkscited by 2,986
  3. 2016WaveNet: A Generative Model for Raw Audiocited by 1,722
  4. 2017Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictionscited by 804
  5. 2020FastSpeech 2: Fast and High-Quality End-to-End Text to Speechcited by 717
  6. 2013Generating Sequences With Recurrent Neural Networkscited by 554
  7. 2017Tacotron: Towards End-to-End Speech Synthesiscited by 490
  8. 2015Attention-Based Models for Speech Recognitioncited by 401
  9. 2018WaveGlow: A Flow-based Generative Network for Speech Synthesiscited by 309
  10. 2020Glow-TTS: A Generative Flow for Text-to-Speech via Monotonic Alignment Searchcited by 273
  11. 2018Efficient Neural Audio Synthesiscited by 219
  12. 2017Parallel WaveNet: Fast High-Fidelity Speech Synthesiscited by 215
11 more
  1. 2020End-to-End Adversarial Text-to-Speechcited by 77
  2. 2018ClariNet: Parallel Wave Generation in End-to-End Text-to-Speechcited by 70
  3. 2017Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attentioncited by 57
  4. 2020Flowtron: an Autoregressive Flow-based Generative Network for Text-to-Speech Synthesiscited by 54
  5. 2021One TTS Alignment To Rule Them Allcited by 47
  6. 2019Location-Relative Attention Mechanisms For Robust Long-Form Speech Synthesiscited by 44
  7. 2021Parallel Tacotron 2: A Non-Autoregressive Neural TTS Model with Differentiable Duration Modelingcited by 31
  8. 2020AlignTTS: Efficient Feed-Forward Text-to-Speech System without Explicit Alignmentcited by 20
  9. 2021On Biasing Transformer Attention Towards Monotonicitycited by 5
  10. 2021Reinforce-Aligner: Reinforcement Alignment Search for Robust End-to-End Text-to-Speechcited by 4
  11. 2020FeatherTTS: Robust and Efficient attention based Neural TTScited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 0; 2026: 1.

3 papers.

  1. 2024Small-E: Small Language Model with Linear Attention for Efficient Speech Synthesiscited by 6
  2. 2024Utilizing Neural Transducers for Two-Stage Text-to-Speech via Semantic Token Predictioncited by 1
  3. 2026Deterministic Prompting for Speaker-Stable Low-Resource Greek TTScited by 0