---
id: deepfilternet
version: "0.5.6"
license: MIT
license_treatment: permissive
maintenance: dormant
---

# deepfilternet — Noise supression using deep filtering

License: permissive · Maintenance: dormant · Downloads: 78.6K/mo

## What it is and what it does

DeepFilterNet is a speech enhancement framework that uses deep learning to suppress noise from audio at 48kHz sampling rate. It wraps a Rust-based STFT/ISTFT processing core with a Python interface, offering both a command-line tool and a programmatic API for batch or real-time noise reduction. The package depends on numpy for numerical work, loguru for logging, requests for downloads, packaging for version handling, sympy for symbolic math, appdirs for configuration paths, and deepfilterlib for core filtering logic.

You can use it to clean up noisy recordings by passing audio files through a pretrained model, or integrate it into Python applications via the init_df() and enhance() functions. The framework supports multiple model variants and includes optional postfiltering for aggressive noise attenuation. Training and dataset preparation are documented but require additional dependencies not bundled in the base wheel.

Use it for:

- Clean up noisy voice recordings or podcast audio before publishing or further processing.
- Suppress background noise in real-time video conferencing or streaming applications.
- Preprocess speech data for downstream tasks like transcription or speaker recognition.
- Batch denoise archived audio collections or archival speech datasets.
- Integrate noise reduction into assistive audio devices via the LADSPA plugin.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

DeepFilterNet removes background noise from full-band audio at 48kHz using deep learning models, providing both command-line and Python API interfaces for speech enhancement.

Yes, if you need robust speech denoising at 48kHz and can tolerate dormant maintenance. The package is stable, well-documented, has no known vulnerabilities, and carries a permissive MIT license. Install friction is low. Verify external deep learning dependencies and model availability before production use.

## Install

pip install deepfilternet
uv add deepfilternet
poetry add deepfilternet

## Installing deepfilternet

Before you install: Low install friction; pure Python wheel with standard dependencies. Maintenance is dormant—last release was 2023-08-31 and last commit 2024-10-17, so expect no active bug fixes or feature updates, though the codebase remains stable.

License in practice: MIT license is permissive; you may use, modify, and distribute this package freely in commercial or private projects with minimal restrictions.

Quickstart:

pip install deepfilternet

from deepfilternet import enhance, init_df

model, df_state, _ = init_df()
enhanced_audio = enhance(model, df_state, noisy_audio)

External deep learning framework must be installed separately; pretrained models are loaded at runtime and may require network access.

Verify before relying:

- Whether pretrained models are bundled with the wheel or must be downloaded separately at runtime.
- GPU acceleration support and whether external deep learning dependencies are pinned or flexible.
- Real-time latency and CPU/memory footprint for typical audio streams.
- Exact PyTorch version compatibility and whether it must be installed before deepfilternet.

## Package facts

- License: MIT (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: dormant
- Downloads: 78.6K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

noise suppression deep learning, speech enhancement neural network, audio denoising 48khz, background noise removal, speech clarity improvement, noise filtering framework, real-time audio denoising, audio-processing, speech-enhancement, deep-learning

[View on SkillFed](https://skillfed.io/packages/deepfilternet) · [View on PyPI](https://pypi.org/project/deepfilternet/)