BLeRIOT-B corpus: in-flight cockpit noises, acoustic impulse responses of cockpits, and electronic responses of aircraft audio processors and CVRs

Aircraft Cockpit Voice Recorders (CVRs) contain sensitive data which are incompatible with open database to develop specific speech processing tools. The access to aircraft is also difficult and expensive. For these reasons, no realistic CVR database exists to be shared with the scientific community to train their algorithms. Acoustic and electronic measurement campaigns were conducted in aircraft cockpits and manufacturer audio test benches to identify impulse responses of the transmission chain from the pilot's voice to the CVR audio files, as well as recordings of flight noises at different flight phases. This led to a model to transform raw speech into CVR-like speech, shared to the community via a Python program and the present open-source database of 1) impulse responses of cockpits, aircraft microphones, aeronautical headset microphones, aircraft audio processors and CVRs, 2) in-flight noises from different flight phases.