dfnstream-py 0.2.0

Real-time audio denoising using DeepFilterNet with streaming

DeepFilterNet Streaming

A Python library for real-time audio denoising using DeepFilterNet, providing streaming audio processing capabilities through C API bindings.

Platform Support

Currently supports the following platforms:

• macOS ARM64 (Apple Silicon)
• Linux x86_64
• Linux ARM64 (aarch64)

Installation

pip install dfnstream-py

Quick Start

from dfnstream_py import DeepFilterNetStreaming
import numpy as np

# Initialize the processor
processor = DeepFilterNetStreaming()

# Process audio chunks
audio_chunk = np.random.randn(1024).astype(np.float32)  # Your audio data
denoised_chunk = processor.process_chunk(audio_chunk)

# Don't forget to cleanup
processor.close()

Examples

The examples/ folder contains sample scripts demonstrating different use cases:

Single File Processing

uv run python examples/enhance_wav.py input_audio.wav

Processes a single WAV file and saves the denoised version as input_audio_denoised.wav.

Batch Directory Processing

uv run python examples/enhance_dir.py /path/to/wav/files/

Processes all WAV files in a directory and saves enhanced versions to an outputs/ folder.

API Reference

DeepFilterNetStreaming

The main class for audio processing.

DeepFilterNetStreaming(
    model_path=None,          # Optional custom model path
    atten_lim=None,           # Attenuation limit in dB (None = no limit)
    log_level="warn",         # Logging level: "error", "warn", "info", "debug", "trace"
    compensate_delay=True,    # Enable delay compensation
    post_filter_beta=0.0      # Post-filter beta (0.0 = disabled)
)

Key Methods

• process_chunk(audio_chunk) - Process a chunk of audio data
• process_frame(audio_frame) - Process a single frame
• finalize() - Get remaining delayed samples
• close() - Clean up resources