# Umamusume DeBERTA-VITS2 TTS 

👌 **Currently, ONLY Japanese is supported.** 👌

💪 **Based on [Bert-VITS2](https://github.com/fishaudio/Bert-VITS2), this work tightly follows [Akito/umamusume_bert_vits2](https://huggingface.co/spaces/AkitoP/umamusume_bert_vits2), from which the Japanese text preprocessor is provided.** ❤

✋ **Please do NOT enter a really LOOOONG sentence or sentences in a single row. Splitting your inputs into multiple rows makes each row to be inferenced separately.** ✋

✋ **请不要在一行内输入长文本，模型会将每行的输入视为一句话进行推理。请将多句话分别放入不同的行中来减少推理时间.** ✋

## Training Details - For those who may be interested

🎈 **This work switches [cl-tohoku/bert-base-japanese-v3](https://huggingface.co/cl-tohoku/bert-base-japanese-v3) to [ku-nlp/deberta-v2-base-japanese](https://huggingface.co/ku-nlp/deberta-v2-base-japanese) expecting potentially better performance, and, just for fun.** 🥰

❤ Thanks to **SUSTech Center for Computational Science and Engineering**. ❤ This model is trained on A100 (40GB) x 2 with **batch size 32** in total.

💪 This model has been trained for **1 cycle, 90K steps (=60 epoch),** currently. 💪

📕 This work uses linear with warmup (7.5% of total steps) LR scheduler with  ` max_lr=1e-4`. 📕

✂ This work clips gradient value to 10 ✂.

⚠ Finetuning the model on **single-speaker datasets separately** will definitely reach better result than training on a huge dataset comprising of many speakers. Sharing a same model leads to unexpected mixing of the speaker's voice line. ⚠

### TODO:

📅 Train one more cycle using text preprocessor provided by [AkitoP](https://huggingface.co/AkitoP) with better long tone processing capacity. 📅