README.md

metadata

title: MOSS-TTS-PNY
emoji: 🎙️
colorFrom: gray
colorTo: yellow
sdk: gradio
sdk_version: 6.15.1
app_file: app.py
short_description: Speaker-conditioned TTS with emotion & energy control
python_version: '3.12'
startup_duration_timeout: 1h

MOSS-TTS-PNY

Speaker-conditioned text-to-speech demo using the ZDisket/MOSS-TTS-PNY model — a fine-tuned MOSS-TTS checkpoint with character voice clones from My Little Pony: Friendship is Magic and Team Fortress 2.

Features

• 43 speakers — MLP:FiM characters (Twilight, Pinkie, Rainbow, etc.) and TF2 characters (Heavy, Scout, Soldier, etc.)
• 12 emotions — Neutral, Happy, Sad, Angry, Annoyed, Fearful, Surprised, Calm, Disgusted, Whispering, Nonverbal, Shouting
• Continuous energy control — adjust the energy/intensity of the generated speech (0.0 – 1.0)
• 48 kHz output via iSTFTNet2 TorchScript CUDA vocoder

How it works

1. Enter the text you want to synthesize.
2. Pick a speaker from the dropdown.
3. Choose an emotion and set the energy level.
4. Click Generate to produce audio.

Notes

• Uses trust_remote_code=True to load the custom MOSS-TTS model code.
• transformers==4.55.0 is pinned to prevent gibberish outputs.
• The model downloads ~12 GB of weights at startup.
• Max 30 seconds of audio per generation.