A method for low latency audio watermark embedding buffers samples of an audio stream in a buffer, including previous blocks of audio samples in the audio stream. It computes a perceptual mask from the audio samples in the buffer, generates a watermark signal; and applies the perceptual mask to the watermark signal for the first block to produce a mask-applied watermark signal. It inserts the mask-applied watermark signal into the audio samples of the first block without waiting for a subsequent audio block of samples in the audio stream and outputs watermarked audio of the first block.