---
license: cc-by-nc-4.0
datasets:
- OOPPEENN/56697375616C4E6F76656C5F44617461736574
- amphion/Emilia-Dataset
- litagin/ehehe-corpus
- joujiboi/japanese-anime-speech
language:
- ja
base_model:
- HKUSTAudio/Llasa-1B-Multilingual
pipeline_tag: text-to-speech
---

# Galgame-Llasa-1B-v3

## Overview

This is the version 3 of the Galgame-Llasa-1B, a Text-to-Speech (TTS) model fine-tuned for Japanese. This model is based on [HKUSTAudio/Llasa-1B-Multilingual](https://huggingface.co/HKUSTAudio/Llasa-1B-Multilingual).

## What's New in v3?

The primary improvement in v3 is the **modification of the text normalization process** during training.

This update leads to more consistent and accurate speech synthesis, further improving upon the advances made in v2.

## What's New in v2 (from v1)?

Version 2 was trained on a larger and more diverse dataset, including the original Galgame dataset and other sources.

As a result, v2 offered several key improvements over the original version:

- **Improved Kanji Reading:** The model handled the reading of Kanji characters more accurately.
- **Enhanced Prosody:** The generated speech had more natural intonation and expressiveness.
- **Greater Voice Diversity:** The model could produce a wider range of voice styles than the previous version.

## License

This model is licensed under the **CC-BY-NC-4.0**.