Studio report: Linux audio for multi-speaker natural speech technology

The Natural Speech Technology (NST) project is the UK's flagship research programme for speech recognition research in natural environments. NST is a collaboration between Edinburgh, Cambridge and Sheeld Universities; public sector institutions the BBC, NHS and GCHQ; and companies including Nuance, EADS, Cisco and Toshiba. In contrast to assumptions made by most current commercial speech recognisers, natural environments include situations such as multi-participant meetings, where participants may talk over one another, move around the meeting room, make non-speech vocalisations, and all in the presence of noises from oce equipment and external sources such as trac and people outside the room. To generate data for such cases, we have set up a meeting room/recording studio equipped to record 16 channels of audio from real-life meetings, as well as a large computing cluster for audio analysis. These systems run on free, Linux-based software and this paper gives details oftheir implementation as a case study for other users considering Linux audio for similar large projects.