A system and method for audio splicing (insertion) of an Ad audio stream in the compressed domain, where variable early delivery of the Ad audio stream and variable bit rate are allowed, without creating audio distortion, glitches, or other digital artefacts or errors, in the resultant audio stream is disclosed. The present system and method provides for a splice delay buffer which delays the first five Ad audio frames until transmission of the last frame of the primary audio stream, but before the splice time. Subsequent Ad audio frames are delayed by a fixed amount, where the fixed amount is greater than the frame delay of the primary audio stream, to allow for ease of splice back to the primary audio stream.