A code is embedded into an audio product so as to be processed therewith for recording and/or broadcast and yet be reliably detected while remaining inaudible to human perception. The code is represented by symbols formed from an impulse function having its energy within a specified frequency range. The audio product is analyzed to find segments which can mask the code based on tonality and a minimum signal energy. When the audio product with an embedded code is detected, decoding thereof involves finding candidate code signals which are checked against preset criteria. In particular, each symbol is made of at least two impulse functions with a preset spacing therebetween.