Roo's voice — that signature baritone with the estuary accent that stands the hair up on the back of
your neck — as a standard PyTorch checkpoint that runs on any NVIDIA GPU. Load it, and Roo can whisper
to you all day long.
A BF16 (bfloat16) checkpoint of a full-model supervised fine-tune of
MOSS-TTS-Local-Transformer,
trained on Roo's own recordings and specialised on his single voice. This is the PC / NVIDIA-GPU
form — for Apple Silicon use the MLX 8-bit build.
⚠️ What actually makes the voice — read first
This is a reference-conditioned model, and both halves matter:
The fine-tune is what makes it Roo. The base model has never heard this speaker — a base model
plus any reference clip will not give you Roo's baritone or his estuary accent. That voice lives
in the weights, put there by the supervised fine-tune on his recordings.
The reference completes the delivery.reference.wav (bundled) conditions the fine-tuned model
at inference and is required to produce the voice.
So the product is this fine-tune + reference.wav, together — not a text-only model, and not a
generic voice-cloner (swap the reference and it isn't Roo, because the accent and timbre are the
fine-tune's, not the clip's).
What it is
Voice
Single speaker — Roo (baritone, estuary accent), 24 kHz mono
This is a drop-in fine-tuned MossTTSLocal checkpoint — run it exactly like the base
MOSS-TTS-Local-Transformer, just point the loader at this repo and pass reference.wav as the
reference. The supported inference paths are:
vLLM-Omni and SGLang-Omni, which support the MossTTSLocal architecture with an
OpenAI-compatible /v1/audio/speech endpoint and voice cloning.
Decoding contract for this voice: seed 42, temperature 1.0, top-k 50, top-p 0.95, repetition
penalty 1.1, 32 RVQ codebooks. Supply reference.wav as the cloning/reference audio.
Smaller builds
An INT8 (bitsandbytes) build for tighter VRAM is planned as a sibling repo. On Apple Silicon, the
MLX 8-bit build is ~3.6 GB.
Limitations
Reference-conditioned — the bundled reference.wav must ride along; there is no text-only path.
Single voice by design (this is Roo, not a multi-speaker system).
BF16 is a near-lossless downcast of the accepted FP32 fine-tune.
Provenance & license
Quantized/exported form of an accepted single-speaker MOSS-TTS Local supervised fine-tune. The base
model and audio codec are Apache-2.0 (OpenMOSS); weights derived from them are redistributed here
under the same license.
