snac
Multi-Scale Neural Audio Codec
Decision gist · record as of 2026-08-14
Yes, if you need neural audio compression for language modeling or token-based audio generation and can work with mono audio on GPU. The low install friction and established model zoo make it straightforward to integrate. However, verify the license before production use, and note that maintenance is dormant—expect no active updates, though the package is stable and has no known vulnerabilities.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires CUDA-capable GPU and torch installed; audio input shape is (batch, channels=1, samples); codes is a list of token sequences at different temporal resolutions.
- Low install friction with a pure Python wheel.
- Maintenance is dormant (last commit 2024-11-19, 702 days since release), but the package remains archived=false and has accumulated 774 stars, suggesting stable prior adoption.
License · maintenance · safety
(unclear) — License treatment is unclear—no SPDX identifier or raw license text is available in the package metadata. Verify the actual license before using in proprietary or copyleft-sensitive contexts.
last release 2024-09-11 (702 days) · last repo commit 2024-11-19 · 774 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 272,475 downloads/mo, #8,209 on PyPI
Alternatives
Verify before relying
pip install snac
import torch
from snac import SNAC
model = SNAC.from_pretrained("hubertsiuzdak/snac_32khz").eval().cuda()
audio = torch.randn(1, 1, 32000).cuda()
with torch.inference_mode():
codes = model.encode(audio)
audio_hat = model.decode(codes)- Actual license identifier and terms—metadata shows 'unclear' treatment with no SPDX or raw license field.
- Python version compatibility—requires_python is unspecified in metadata.
- Whether models support stereo or multi-channel audio beyond the documented mono-only constraint.
What it is and what it does
SNAC is a neural audio codec that encodes audio into hierarchical discrete tokens at low bitrate. It follows the architecture of SoundStream, EnCodec, and DAC but uses coarse tokens sampled at lower frequency to cover broader time spans, reducing bitrate while enabling long-context audio modeling—for example, 2048 coarse tokens at ~10 Hz can represent ~3 minutes of audio structure.
The package provides pretrained models for speech (24 kHz, 0.98 kbps) and music/sound effects (32 kHz and 44 kHz, 1.9–2.6 kbps). It depends on torch, numpy, einops, and huggingface-hub to load models. Encoding and decoding are straightforward: load a model, call encode() on audio tensors, and retrieve a list of token sequences at different resolutions.
Use it for
- Compress speech or music into discrete tokens for downstream language model training on audio.
- Reduce audio bitrate for storage or transmission while preserving perceptual quality.
- Generate audio using discrete token sequences in a language modeling framework.
- Reconstruct audio from encoded tokens for real-time or batch audio processing pipelines.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, if you need neural audio compression for language modeling or token-based audio generation and can work with mono audio on GPU.
The low install friction and established model zoo make it straightforward to integrate. However, verify the license before production use, and note that maintenance is dormant—expect no active updates, though the package is stable and has no known vulnerabilities.
Install
snac on PyPI
Before you install
Low install friction with a pure Python wheel. Maintenance is dormant (last commit 2024-11-19, 702 days since release), but the package remains archived=false and has accumulated 774 stars, suggesting stable prior adoption.
Requires CUDA-capable GPU and torch installed; audio input shape is (batch, channels=1, samples); codes is a list of token sequences at different temporal resolutions.
License in practice
License treatment is unclear—no SPDX identifier or raw license text is available in the package metadata. Verify the actual license before using in proprietary or copyleft-sensitive contexts.
Quickstart
pip install snac
import torch
from snac import SNAC
model = SNAC.from_pretrained("hubertsiuzdak/snac_32khz").eval().cuda()
audio = torch.randn(1, 1, 32000).cuda()
with torch.inference_mode():
codes = model.encode(audio)
audio_hat = model.decode(codes)Verify before relying
- Actual license identifier and terms—metadata shows 'unclear' treatment with no SPDX or raw license field.
- Python version compatibility—requires_python is unspecified in metadata.
- Whether models support stereo or multi-channel audio beyond the documented mono-only constraint.
Package facts
| License | Not declared unclear |
| Python support | Not specified |
| Install friction | Low. Pure-Python wheel |
| Runtime dependencies | 4 packagestorchnumpyeinopshuggingface-hub |
| Maintenance | Dormant 702 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 272,475 / month, #8,209 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
Evidence: snac-1.2.1-py3-none-any.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “discrete audio tokens”
- snacSNAC compresses audio into discrete hierarchical tokens at low…
- s3tokenizerConverts audio into discrete semantic speech tokens using a PyTorch…
- encodecEnCodec is a neural audio codec that compresses audio to low bitrates…
Give your agent the search over MCP, or paste the wish link into any chat.
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to call 100+ LLM providers (OpenAI, Anthropic, Gemini, Bedrock, Azure, and others) using OpenAI-compatible API format, available as both a Python SDK and a self-hosted AI Gateway proxy server.
Install it if you need to work with multiple LLM providers or want to centralize LLM routing in your organization.
Client library and CLI tool for downloading, uploading, and managing models, datasets, and repositories on the Hugging Face Hub platform.
Install it if you work with Hugging Face Hub models or datasets.
LangChain provides a framework for building agents and LLM-powered applications by composing language models, tools, and memory through a unified API that abstracts over multiple model providers.
hf-xet provides chunk-based deduplication and efficient file transfer for the Hugging Face Hub, enabling faster uploads and downloads of large files with local disk caching.
Tokenizers converts raw text into token sequences for NLP models, with support for training custom vocabularies and using pre-built tokenizers (BPE, WordPiece) optimized for speed via Rust.
Transformers provides a unified framework for loading, fine-tuning, and running state-of-the-art pretrained models across text, vision, audio, video, and multimodal tasks using PyTorch, JAX, or TensorFlow.
Install it if you need to run or train any transformer-based model for NLP, vision, audio, or multimodal tasks.
See also descript-audio-codec · encodec · laion-clap · panns-inference · vocos · s3tokenizer · silk-python · openunmix · mlx-audio · music-assistant-models