Own voice recognition (OVR) for hearing aids, detects time instances where the person wearing the device is speaking. Classification of the own voice is performed dependent on a fixed or adaptive detection threshold. Automatic tuning in a real-time system depends on general noise statistics in the input signals. The noise is removed from the received signal and is characterized by signal-to-noise ratio and noise color. An optimal detection threshold for own voice recognition is determined based on the noise characteristics. A noise detection model is created by smoothed Voronoi tessellation. Own voice detection is performed by a processor.