ManaTTS Persian: a recipe for creating TTS datasets for lower resource languages

TL;DR

This paper introduces ManaTTS, a large Persian speech dataset, along with tools and methods for dataset creation and speech recognition, enabling high-quality TTS and low-resource language processing.

Contribution

It provides the largest open Persian speech corpus and a transparent pipeline with novel tools for dataset collection and forced alignment tailored for low-resource languages.

Findings

Achieved a MOS of 3.76 with the TTS model, close to natural speech quality.

Developed a fully open, MIT-licensed pipeline for dataset creation and alignment.

Extended speech recognition evaluation with the VirgoolInformal dataset.

Abstract

In this study, we introduce ManaTTS, the most extensive publicly accessible single-speaker Persian corpus, and a comprehensive framework for collecting transcribed speech datasets for the Persian language. ManaTTS, released under the open CC-0 license, comprises approximately 86 hours of audio with a sampling rate of 44.1 kHz. Alongside ManaTTS, we also generated the VirgoolInformal dataset to evaluate Persian speech recognition models used for forced alignment, extending over 5 hours of audio. The datasets are supported by a fully transparent, MIT-licensed pipeline, a testament to innovation in the field. It includes unique tools for sentence tokenization, bounded audio segmentation, and a novel forced alignment method. This alignment technique is specifically designed for low-resource languages, addressing a crucial need in the field. With this dataset, we trained a Tacotron2-based TTS model, achieving a Mean Opinion Score (MOS) of 3.76, which is remarkably close to the MOS of 3.86 for the utterances generated by the same vocoder and natural spectrogram, and the MOS of 4.01 for the natural waveform, demonstrating the exceptional quality and effectiveness of the corpus.