ATC Recognition and Management

Air Traffic Control Speech Recognition and Management

ATC Custom Speech Recognition Development

MatrixHCI can offer you its 8+ years of experience developing aviation and aerospace speech recognition solutions in developing focused speech solution for your unique product needs.  Whether you just need the speech recognition solution or the design of a complete software system, MatrixHCI can assist you and give you access to an instant experienced results-oriented leading edge development and research lab.

Please contact us for a free consultation, and learn more about how we can help meet your current and future product development needs.

ATC Speech Recognition Design

ATC radio transmissions are considered one of the most complex and challenging speech recognition environments and have been considered impossible to recognize by many major speech recognition companies in business today.

MatrixHCI has pushed technology and broken the barrier to ATC speech recognition by applying its many years of speech recognition development in this area.

MatrixHCI has been developing projects in the ATC recognition area for over 8 years now, and has successfully developed methods and intelligent processes that aid in the successful recognition of ATC transmissions.

shutterstock_3395577
Screen Shot 2015-12-03 at 10.18.18 PM

ATC Speech Recognition Challenges

Recognition of ATC radio transmissions and commands are especially challenging for a variety of reasons:

1. Radio Transmission Quality: ATC radio transmission quality can vary from frequency to frequency and also from airport to airport.  Additionally, distance from the aircraft to the actual transmission source of the ATC transmitter can impact audio quality.

2. ATC Speech is Fast and Furious: ATC controllers talk very fast.  Much faster than normal speech patterns used in regular conversation.  As a result standard acoustic models do not work well in this environment.

3. FAA Phraseology: ATC controllers utilize very strict FAA approved phraseology when issuing commands to pilots.  This phraseology requires custom training of acoustic models and language models.  Additionally, the phraseology changes based on flight state.  What is spoken on departure is totally different on approach, etc.

4. Shared Frequency: An ATC frequency is shared by the ATC controller and all aircraft communicating on that frequency.  There are also multiple frequencies for each airport.  Distinguishing between ATC commands and pilot return responses can be problematic.  Additionally, pilots and ATC controllers can decide to key their mics at the same time stepping on each others transmissions.

5. Tail Number Recognition: An ATC controller speaks the aircraft tail number or flight number at the beginning of each ATC command.  It is common for controllers to key their mic at the same time they start speaking causing the first part of the tail number or flight number to be clipped from the audio transmission.

These are just a few audio challenges faced when designing ATC speech recognition.

ATC Speech Recognition Design

MatrixHCI has extended the technology envelope of speech recognition and has proved that it is possible to recognize ATC radio commands successfully. With over 8 years of experience designing and developing intelligent ATC-based recognition processes, MatrixHCI has delivered speech recognition solutions that deliver proven recognition results in the aviation and ATC arenas.

Because MatrixHCI developers also have the advantage of having licensed pilot experience, our understanding of ATC commands and flight space, and rules and regulations, are extensive.  This allows us to design systems that work from the pilot’s point of view.  Often a pilot may head a command from ATC but not fully understand the command given due to clipped transmissions, diversion of attention on other tasks on the flight deck, or because the controller delivers the command too fast.

When this happens, pilots use their knowledge about their flight state, position and the state of the aircraft, and the prior ATC command to “fill in the gaps” and build up an understanding of what the ATC controller said.

At MatrixHCI, we apply some of the same techniques used by human pilots to recognize, fill gaps, and build up a complete understanding of what is said. This is done by incorporating special AI-based techniques and special custom developed acoustic models and language models all working in concert to produce desired ATC recognition results.

MatrixHCI also uses other types of data collection available to aircraft while in flight to further increase recognition understanding and accuracy, such as ADS-B data.  Additional data sources are also being incorporated into the process as ongoing research and development continues.

shutterstock_312836912
IMG_8409
IMG_5925
IMG_4904

ATC Audio Data Collection and Transcription

MatrixHCI is constantly collecting ATC transmission data from airports throughout the US.  Because ATC speech recognition accuracy also has a direct correlation to the amount of audio that has been sampled and trained into the acoustic model, MatrixHCI is building a massive audio repository of ATC transmissions from all phases of flight on many frequencies throughout the geographic areas of the US.

Additionally, because ATC controllers work on 20 minute shifts and repeat their shifts during the days and weeks, it is problematic to acquire a wide range of different audio samples from just a single airport or two.  To compound the issue, there are a smaller percentage of female ATC controllers, and the amount of audio samples from female ATC controllers is harder to acquire.

Also, each geographic region of the country has different language accents and inflections that need to be factored into a successful ATC speech recognition system.

To solve these problems of data collection, MatrixHCI is making a major investment and push into collecting a large and diverse set of ATC audio corpora that can be used in building specialized speech recognition solutions that involve ATC transmissions.

Data has been collected or collection is planned in the following airports in 2015/2016:

-KLAX (Los Angeles International Airport)
-KSLC (Salt Lake International Airport)
-KLAS (McCarran International Airport) in Las Vegas
-KDFW (Dallas/Fort Worth International Airport)
-KATL (Hartsfield-Jackson Atlanta International Airport)
-KORD (Chicago O’Hare International Airport)
-KBOS (Boston Logan International Airport)
-KJFK (John F. Kennedy International Airport)
-KEWR (Newark Liberty International Airport)
-KMIA (Miami International Airport)
-KDEN (Denver International Airport)
-KSFO (San Francisco International Airport)
-KSEA (Seattle-Tacoma International Airport)

Additionally, it has been proven that ATC controllers speak differently and issue commands in modified ways whenever air traffic in their area is overly intense and congested.  MatrixHCI has also collected hundreds of hours of ATC audio under stressed situations.

Recently, MatrixHCI collected over 200 hours of stressed ATC audio during the recent 2015 NBAA (National Business Aircraft Association) trade show in Las Vegas.  This show is the 4th largest trade show in the US, and air traffic at McCarran International Airport is at its highest during this week.  We positioned our mobile 40 ft RV audio collection unit at the airport during this show and captured audio on three audio channels each scanning multiple ATC frequencies, encompassing tower, approach, departure, and ground transmissions.

Other airports are planned in the future, and we are also willing to collect ATC and other FAA frequency audio for your unique needs using our custom mobile 40 ft RV audio collection bus.

For more information on our audio collection capabilities and our mobile audio collection vehicle, please contact us.

MatrixHCI offers Custom Software Development with an emphasis on Advanced Cutting-Edge Speech Recognition.

If you have a need for specialized custom speech recognition solutions, please contact us for a free confidential consultation.