spark-tts-genshin-charactors

Spark TTS finetuned in genshin charactors voices.

• github code: https://github.com/nonwesjoe/genshin-sparktts
• kaggle notebook: https://www.kaggle.com/code/suziwsz/genshin-sparktts/

Available charactors

• paimon, hutao, furina, kazuha, xiao, mona, ganyu, xiangling, shotgun, citlali, barbara, zhongli, venti, nahida, kaeya, yaoyao, yoimiya, nilou.(each charactor in one full finetuned model)

Usage

• python 3.12 suggested
• git clone https://github.com/nonwesjoe/genshin-sparktts.git && cd genshin-sparktts
• when cuda is availabel, install torch 2.7.1 on cuda <code>pip install torch torchaudio torchvision -i https://download.pytorch.org/whl/cu118/</code>
  else, install torch 2.7.1 on cpu <code>pip install torch torchaudio torchvision -i https://download.pytorch.org/whl/cpu</code>
• install other requirements <code>pip install -r requirements.txt</code>
• in terminal set some environment variables

export CHARACTOR=nahida                                         # or other charactors  
export MODEL_PATH=/kaggle/working/genshin/                      # your model path  
export INPUT_TEXT="楼下发荔枝了吗?那我们快去领取！"  # text to be converted  

• download model files: defaultly, download one specific charactor model set in environment variable CHARACTOR. model will be download in ./genshin <code>

python3 download.py </code>

• run code to convert text to audio. audio outputs sparktts.wav. <code>

python3 run.py </code>

Detail

• this model is trianed on float32 but saved as float16 for less VRAM and Storage usage.

Example

▶ Furina(芙宁娜) play
▶ Kazuha(万叶) play
▶ Paimon(派蒙) play
▶ Hutao(胡桃) play
▶ Xiao(魈) play
▶ Citlali(茜特菈莉) play

Acknowledgement

• Spark-TTS
• Genshin dataset
• Unsloth