# Scalable Multilingual Frontend for TTS

arXiv: https://arxiv.org/abs/2004.04934 · HTML: https://mlsanity.com/paper/2004.04934/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 10 Apr 2020.

## Built on by (most cited first, 8 of 8)

- [Proteno: Text Normalization with Limited Data for Fast Deployment in Text to Speech Systems](https://mlsanity.com/paper/2104.07777/) (2021)
- [Review of end-to-end speech synthesis technology based on deep learning](https://mlsanity.com/paper/2104.09995/) (2021)
- [Transformer-based Models of Text Normalization for Speech Applications](https://mlsanity.com/paper/2202.00153/) (2022)
- [Multilingual context-based pronunciation learning for Text-to-Speech](https://mlsanity.com/paper/2307.16709/) (2023)
- [Compact Neural TTS Voices for Accessibility](https://mlsanity.com/paper/2501.17332/) (2025)
- [Multi-Task Learning for Front-End Text Processing in TTS](https://mlsanity.com/paper/2401.06321/) (2024)
- [Acquiring Pronunciation Knowledge from Transcribed Speech Audio via Multi-task Learning](https://mlsanity.com/paper/2409.09891/) (2024)
- [Positional Description for Numerical Normalization](https://mlsanity.com/paper/2408.12430/) (2024)

## Abstract (arXiv)

This paper describes progress towards making a Neural Text-to-Speech (TTS) Frontend that works for many languages and can be easily extended to new languages. We take a Machine Translation (MT) inspired approach to constructing the frontend, and model both text normalization and pronunciation on a sentence level by building and using sequence-to-sequence (S2S) models. We experimented with training normalization and pronunciation as separate S2S models and with training a single S2S model combining both functions. For our language-independent approach to pronunciation we do not use a lexicon. Instead all pronunciations, including context-based pronunciations, are captured in the S2S model. We also present a language-independent chunking and splicing technique that allows us to process arbitrary-length sentences. Models for 18 languages were trained and evaluated. Many of the accuracy measurements are above 99%. We also evaluated the models in the context of end-to-end synthesis against our current production system.