BC Stage drum generator — Stable Audio Open Small (ONNX)
On-device drum one-shot generation for the BC Stage app (BC Labs).
This is an ONNX conversion of
Stable Audio Open Small.
See NOTICE for exactly what was changed.
Powered by Stability AI. Licensed under the Stability AI Community License (LICENSE).
Files
| file | what |
|---|---|
dit_fp16_ext.onnx + dit_fp16_ext.bin |
diffusion transformer, float16 weights, external data |
cozucu.onnx |
autoencoder decoder (latent → 44.1 kHz stereo) |
istemler.bin / istemler.json |
precomputed prompt conditioning + metadata, base package (kick, snare, closed hi-hat, styles) |
dit_perk_fp16_ext.onnx + dit_perk_fp16_ext.bin |
fine-tuned diffusion transformer: percussion, tom set, open/half/pedal hi-hat, rimshot/brush snare (float16) |
istemler_perk.bin / istemler_perk.json |
precomputed prompt conditioning + metadata for the percussion package |
The percussion package was fine-tuned on CC BY / CC0 recordings; see NOTICE for attributions.
Inputs
DiT: x (1,64,43), t (1), cross (1,65,768), glob (1,768) → v (1,64,43).
Decoder: z (1,64,43) → (1,2,88064).
Sampler: 8-step ping-pong, t = sigmoid(-linspace(-6, 2, 9)), t[0]=1, t[8]=0;
d = x - t_i·v, x = (1 - t_{i+1})·d + t_{i+1}·noise.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for Cekergil/bc-stage-drums
Base model
stabilityai/stable-audio-open-small