Audio processing device and a method for estimating a signal-to-noise-ratio of a sound signal

An audio processing device, comprising

• at least one input unit for providing a time-frequency representation Y(k,n) of an electric input signal representing a time variant sound signal consisting of target speech signal components S(k,n) from a target sound source TS and noise signal components N(k,n) from other sources than the target sound source, where k and n are frequency band and time frame indices, respectively,
• a noise reduction system configured to
• determine a first signal to noise ratio estimate γ(k,n) of said electric input signal,
• determine second signal to noise ratio estimate ζ(k,n) of said electric input signal from said first signal to noise ratio estimate γ(k,n) based on a recursive algorithm comprising a recursive loop, and to
• determine said second signal to noise ratio estimate γ(k,n) by non-linear smoothing of said first signal to noise ratio estimate γ(k,n), or a parameter derived therefrom, and wherein said non-linear smoothing is controlled by one or more bias and/or smoothing parameters;
• and
• wherein a determination of said one or more bias and/or smoothing parameters comprises the use of supervised learning implementing one or more trained neural networks.