Showing 1–1 of 1 results for author: Sainburg, T

Noisereduce: Domain General Noise Reduction for Time Series Signals

Authors: Tim Sainburg, Asaf Zorea

Abstract: Extracting signals from noisy backgrounds is a fundamental problem in signal processing across a variety of domains. In this paper, we introduce Noisereduce, an algorithm for minimizing noise across a variety of domains, including speech, bioacoustics, neurophysiology, and seismology. Noisereduce uses spectral gating to estimate a frequency-domain mask that effectively separates signals from noise… ▽ More Extracting signals from noisy backgrounds is a fundamental problem in signal processing across a variety of domains. In this paper, we introduce Noisereduce, an algorithm for minimizing noise across a variety of domains, including speech, bioacoustics, neurophysiology, and seismology. Noisereduce uses spectral gating to estimate a frequency-domain mask that effectively separates signals from noise. It is fast, lightweight, requires no training data, and handles both stationary and non-stationary noise, making it both a versatile tool and a convenient baseline for comparison with domain-specific applications. We provide a detailed overview of Noisereduce and evaluate its performance on a variety of time-domain signals. △ Less

Submitted 19 December, 2024; originally announced December 2024.

Comments: Python library: https://github.com/timsainb/noisereduce or `pip install noisereduce`