In this work, we present a novel robust audio watermarking method in wavelet domain. Embedding of watermarks is performed in a perceptually significant set of sub-bands of the host audio. In order to enhance security, watermark is embedded in a randomly selected sub-band of the host audio according to a secret key. Watermark embedding to a mapped subband is done in a predefined time period, similarly to frequency hopping approach in digital communications. Detection method is a statistical method motivated by modified patchwork algorithm, relying on large data sets. Using simple Haar decomposition and test statistics, the algorithm proved itself in the terms of computational complexity. Subjective listening test proved high perceptual transparency of the watermarked audio clips. Using embedding rate of 4.61 bps, high robustness against common audio watermarking attacks was obtained, with average bit error rate lower than 1%.