SAI - Cricket (Audio & Speech Annotation)
In this project, I performed high-precision audio annotation and transcription to improve the performance of virtual assistant models (Siri). My work involved analyzing complex, multi-speaker interactions in Norwegian (nb_NO) to identify and segment 'User,' 'Siri,' and 'Bystander' turns. Using a specialized time-series spectrogram tool, I applied precise timestamps to acoustic events including voice triggers (Invocations), fillers, and pauses.A key component of this role was performing high-quality transcription using Post-Inverse Text Normalization (Post-ITN) standards, ensuring that spoken Norwegian was converted into a screen-ready format. I also managed complex linguistic edge cases, such as dialect-to-Bokmål conversion, compound word structures, and truncated/unclear speech tagging. This project required strict adherence to evolving data security (InfoSec) protocols and meticulous attention to detail to ensure the accuracy of the training datasets." A key component of this role was performing high-quality transcription using Post-Inverse Text Normalization (Post-ITN) standards, ensuring that spoken Norwegian was converted into a screen-ready format. I also managed complex linguistic edge cases, such as dialect-to-Bokmål conversion, compound word structures, and truncated/unclear speech tagging.