DATA LABELLER
The TELUS SONIC project involved audio data labeling at scale to improve speech recognition and conversational AI systems. The project scope required reviewing and annotating various audio data types. The data types included various accents, speech clarity levels, and noise conditions. The project aimed to improve the robustness of speech recognition and conversational systems. The project activities included audio transcription, speech segmentation, speaker identification, and labeling non-speech events such as noise, silence, and speech overlaps. The activities were performed according to strict SONIC annotation guidelines. The project activities involved handling high-volume data with thousands of audio files. The activities were performed within daily productivity and accuracy requirements. The quality indicators included adherence to strict labeling guidelines, multi-stage quality assurance activities, inter-annotator agreements, spot audits, and continuous feedback.