Originally Posted by Coagie
The guys in the lab can take unmolested, digitized factory copies of each standard Airbus alarm voice or sound, make it 180 degrees out of phase (make the positive voltage negative and vice versa) and sum each with each corresponding alarm on the CVR recording, increasing the amplitude until each alarm sound is at zero amplitude. This should leave the remaining sounds, including the pilots voices, easier to hear and understand.
This could take a day or two
Fortunately there is already ready-made sophisticated audio spectral editing software for this, e.g, Izotope RX4. It easily allows a skilled forensic analyst to null out specific spectral patterns such as alarms which obscure speech.