SNR estimation for clipped audio based on amplitude distribution

Waveform clipping is a common problem in digital audio signal processing. Clipping changes the audio amplitude distribution and creates noise, which results in the decrease of audio signal-to-noise ratio (SNR). To estimate the degree of the clipping effect, we propose a method for SNR estimation of clipped signals base on the audio amplitude distribution. We first locate the clipping intervals of audio signals with high precision based on the differences between the amplitude probability density functions (PDFs) of normal and clipped audio signals. Then, by using the Gamma distribution, the clipped amplitude PDF is restored to estimate the non-clipped amplitude distribution. Finally, based on the restored PDF, the estimated SNR of distorted audio signals are calculated, which can indicate the degree of the clipping effect. Experiments on 3200 audios show that our approach can achieve a low error of 0.017 in the clipped values locating, and an error less than 0.5dB in the SNR estimation.