Images from the moon landing and other major space moments of the 1960s are familiar to many Americans. But what about the sounds? NASA recorded thousands of hours of audio tapes from the Apollo missions that have sat unheard in storage for decades.
John Hanson, the associate dean of research at the University of Texas at Dallas, is part of a team trying to reconstruct that massive archive of space history for ears in the 21st century in the project, Explore Apollo.
Interview Highlights
On how the project started: When we think about NASA, we think about how the astronauts accomplished an enormous feat in walking on the moon. But one of the things of that people tend to not think about as much is the thousands of people behind the scenes, working collaboratively to make sure that the mission was going to be a success. This project is really focused on trying to understand how people work collaboratively together to achieve these really challenging engineering and scientific obstacles.
On how they transcribed the audio: In each of these tapes, there are 30 tracks that are being recorded simultaneously and many of these tracks are loops. There are multiple people, anywhere from three to as many as 35 people, that may be speaking on a particular track. Now, NASA is really organized and so are the engineers and scientists here. You hear these tones called "Quindar tones" (the high-pitched beeps) and that's what NASA used to ensure that when the astronauts speak, everyone else has to be quiet.
Other than that, you have lots people all working collaboratively to make sure that the rocket is heading in the right direction, that there is enough fuel, enough oxygen, vital signs for the astronauts. Constantly, there are people monitoring all these factors. The technologies we had to develop required, first, that we identify where there is speech and where there is silence, and when there is speech recognizing what is being said, and how many people might be speaking at the time. Those are some of the challenges.
On the device they built to access the tapes: The playback system only provides support for one track to be played at a time, so if we were just looking at Apollo 11, it would have taken close to about 170 years of continuously digitizing for us to just get Apollo 11. So, we worked with a company to help design a new 30-track read head, a small, magnetic device that senses analog sound that's recorded on the 1-inch tape. I personally installed it myself on the system — a big, big fear because if I missed up, we'd be in trouble; we wouldn't be able to get anything off.
On pushing the boundaries of transcription: In order to get that text, you really need to understand what NASA's trying to accomplish. They're trying to get as much audio in the shortest amount of time, which means they use an acronym for everything. For us to build just the language model to do the speech recognition, we couldn't use Siri or commercial-based speech recognition packages you might buy because none of them would have the vocabulary or lexicon that NASA would use. So we drew from thousands of books and manuscripts from NASA. We pulled close to 4.5 billion words from NASA archives of text in order to build a language model that would characterize the types and words that they would use for each of the Apollo missions. Without that, the speech recognition system would never be able to work on this data.
This interview has been edited for clarity and length.
