speexdsp-ns-vulcanlabs 0.1.3

Python bindings of speexdsp noise suppression library (Vulcanlabs fork with multi-platform wheels)

speexdsp-ns-vulcanlabs

Python bindings for SpeexDSP noise suppression library with pre-built wheels for multiple platforms.

This is a fork of TeaPoly/speexdsp-ns-python with added support for:

• ✅ Linux x86_64
• ✅ Linux ARM64 (aarch64)
• ✅ macOS x86_64 (Intel)
• ✅ macOS ARM64 (Apple Silicon M1/M2/M3)
• ✅ Python 3.8 - 3.12

Installation

pip install speexdsp-ns-vulcanlabs

Usage

from speexdsp_ns import NoiseSuppression

# Create noise suppression instance
# frame_size: number of samples per frame (e.g., 256)
# sample_rate: audio sample rate in Hz (e.g., 16000)
ns = NoiseSuppression.create(frame_size=256, sample_rate=16000)

# Process audio frame (must be bytes of int16 samples)
# Input: raw PCM audio bytes (frame_size * 2 bytes for int16)
# Output: noise-suppressed audio bytes
processed_audio = ns.process(raw_audio_bytes)

Example: Process WAV file

import wave
from speexdsp_ns import NoiseSuppression

frame_size = 256

# Open input file
with wave.open('input.wav', 'rb') as infile:
    sample_rate = infile.getframerate()
    
    # Create noise suppression
    ns = NoiseSuppression.create(frame_size, sample_rate)
    
    # Open output file
    with wave.open('output.wav', 'wb') as outfile:
        outfile.setnchannels(1)
        outfile.setsampwidth(2)  # 16-bit
        outfile.setframerate(sample_rate)
        
        while True:
            data = infile.readframes(frame_size)
            if len(data) != frame_size * 2:
                break
            
            # Process frame
            processed = ns.process(data)
            outfile.writeframes(processed)

Requirements

• Python 3.8+
• No additional system dependencies required (pre-built wheels include speexdsp)

Building from source

If you need to build from source:

# Linux (Ubuntu/Debian)
sudo apt-get install libspeexdsp-dev swig
pip install .

# macOS
brew install speexdsp swig
export LDFLAGS="-L$(brew --prefix speexdsp)/lib"
export CPPFLAGS="-I$(brew --prefix speexdsp)/include"
pip install .

License

BSD-3-Clause License (same as original speexdsp-ns-python)

Credits

• Original implementation: TeaPoly/speexdsp-ns-python
• SpeexDSP library: xiph/speexdsp
• Multi-platform wheels: Vulcanlabs