cudabenchmarktest/stable-audio-open-1.0

stable-audio-open-1.0

Generates variable-length (up to 47 s) stereo audio at 44.1 kHz from text prompts. Latent diffusion architecture with an autoencoder, a T5 text encoder, and a transformer-based diffusion (DiT) operating in the autoencoder's latent space.

This repository is an unmodified redistribution of `stabilityai/stable-audio-open-1.0`. Weights, configs, license, and dataset attribution files are preserved verbatim.

Files

• —model.safetensors (~4.85 GB) — primary weights.
• —model.ckpt (~4.85 GB) — same weights in .ckpt format for stable_audio_tools.
• —model_config.json, model_index.json — pipeline configs.
• —LICENSE.md — Stability AI Community License (verbatim).
• —fma_dataset_attribution2.csv, freesound_dataset_attribution2.csv — training-data attribution (required by the license).

Inference

import torch
import torchaudio
from einops import rearrange
from stable_audio_tools import get_pretrained_model
from stable_audio_tools.inference.generation import generate_diffusion_cond

device = "cuda" if torch.cuda.is_available() else "cpu"
model, model_config = get_pretrained_model("cudabenchmarktest/stable-audio-open-1.0")
sample_rate = model_config["sample_rate"]
sample_size = model_config["sample_size"]
model = model.to(device)

conditioning = [{"prompt": "128 BPM tech house drum loop", "seconds_start": 0, "seconds_total": 30}]
output = generate_diffusion_cond(
    model, steps=100, cfg_scale=7, conditioning=conditioning,
    sample_size=sample_size, sigma_min=0.3, sigma_max=500,
    sampler_type="dpmpp-3m-sde", device=device,
)
output = rearrange(output, "b d n -> d (b n)")
output = output.to(torch.float32).div(torch.max(torch.abs(output))).clamp(-1, 1).mul(32767).to(torch.int16).cpu()
torchaudio.save("output.wav", output, sample_rate)

License and attribution

Governed by the Stability AI Community License Agreement (see LICENSE.md). Permits research, non-commercial use, and commercial use for organizations or individuals with less than $1M USD in total annual revenue. Above that threshold a separate Stability Enterprise license is required.

Training-data attribution: see the FMA and Freesound CSV files. Distribution of these attribution files alongside the weights is a license requirement and is preserved here.

• —Original release: Stability AI (stabilityai/stable-audio-open-1.0).
• —This redistribution: weights and configs unmodified, LICENSE preserved, README replaced. No additional modifications.