Natural Language Audio Rater
As part of a data training initiative with LXT, I was responsible for evaluating and labeling text-to-speech (TTS) and generative audio samples in English. My primary role involved assessing the naturalness, clarity, pronunciation, and overall human-likeness of AI-generated audio in comparison to human-spoken recordings. This included selecting the most accurate and natural-sounding output from multiple candidates. The project required a fine-tuned ear for phonetic detail and a deep understanding of English linguistic variation. I reviewed samples across a wide range of English accents—including American, British, and international varieties—and evaluated content on diverse topics such as music, geography, pop culture, and casual conversation. In addition to selection and labeling, I provided detailed feedback on prosody, enunciation, emotional tone, and cadence to improve voice model training. This work directly contributed to refining speech synthesis systems, enhancing the quality