Grammar Based Speaker Role Identification for Air Traffic Control Speech Recognition

Kurzfassung

Automatic Speech Recognition (ASR) for air traffic control is generally trained by pooling Air Traffic Controller (ATCO) and pilot data. In practice, this is motivated by the proportion of annotated data from pilots being less than ATCO’s. However, due to the data imbalance of ATCO and pilot and their varying acoustic conditions, the ASR performance is usually significantly better for ATCOs speech than pilots. Obtaining the speaker roles requires manual effort when the voice recordings are collected using Very High Frequency (VHF) receivers and the data is noisy and in a single channel without the push-totalk (PTT) signal. In this paper, we propose to (1) split the ATCO and pilot data using an intuitive approach exploiting ASR transcripts and (2) consider ATCO and pilot ASR as two separate tasks for Acoustic Model (AM) training. The paper focuses on applying this approach to noisy data collected using VHF receivers, as this data is helpful for training despite its noisy nature. We also developed a simple yet efficient knowledgebased system for speaker role classification based on grammar defined by the International Civil Aviation Organization (ICAO). Our system accepts as input text, thus, either gold annotations or transcripts generated by an ABSR system. This approach provides an average accuracy in speaker role identification of 83%. Finally, we show that training AMs separately for each task, or using a multitask approach, is well suited for the noisy data compared to the traditional ASR system, where all data is pooled together for AM training.