ponybook-tts

Ponybook TTS — 58-voice audiobook model

The TTS model behind Ponybook, a self-hosted audiobook reader that casts book characters to voices automatically. 58 speakers: the My Little Pony cast (mane six, princesses, students, side characters), the Team Fortress 2 cast (useful stand-ins when no pony fits a gruff or adult character), and a dedicated Anon voice for second-person fics. 12-class emotion conditioning with continuous energy and strength controls.

This is a fan-made fine-tune derived from ZDisket/MOSS-TTS-PNY (itself built on OpenMOSS MOSS-TTS), extended with additional characters and a delivery/pronunciation adapter. Voices are trained from show audio: non-commercial, personal use.

Contents

Not included (unmodified, pulled from the base repo by the Ponybook setup script): the MOSS audio codec (moss_audio_tokenizer/) and the iSTFTNet vocoder — download them from ZDisket/MOSS-TTS-PNY.

Use

Don't wire this up by hand — use the Ponybook repo's tts-server/: setup.sh downloads everything into place and run_server.sh serves a simple HTTP TTS API (FastAPI: /tts, /speakers, /health) on an NVIDIA GPU (~9–12 GB VRAM default preset, lighter presets available).

Speaker ids are positional: speaker_id_map.csv and the checkpoint's embedding table must always travel together, and the voice_patch/ adapter matches this exact checkpoint.

Roster

Ids 0–32 (MLP main cast): Apple Bloom, Applejack, Autumn Blaze, Big Mac, Cadance, Celestia, Cheerilee, Chrysalis, Cozy Glow, Diamond Tiara, Discord, Fluttershy, Goldie Delicious, Granny Smith, Luna, Maud, Mrs. Cake, Pinkie, Rainbow, Rarity, Scootaloo, Shining Armor, Spike, Spitfire, Starlight, Sunburst, Sunset Shimmer, Sweetie Belle, Thorax, Tirek, Trixie, Twilight, Zecora. Ids 33–45 (TF2): Demo, Engineer, Heavy, Medic, Merasmus, Pauling, Scout, Sniper, Soldier, Spy, Saxton Hale, Administrator, Pyro. Ids 46–56 (MLP extras): Smolder, Quibble Pants, Silverstream, Sandbar, Yona, Gallus, Ember, Ocellus, AK Yearling, Gilda, Derpy. Id 57: Anon.