Worked on enhancing audio quality in the davideuler/index-tts repository by developing a feature to trim ultra-long silences in text-to-speech outputs. Focused on improving the naturalness and perceived quality of generated speech, the approach involved implementing a remove_long_silence function that detects and removes excessive consecutive silence tokens from audio codes. This solution addressed the issue of unnatural audio gaps in synthesized speech, resulting in smoother and more realistic output. The work leveraged Python and deep learning techniques, with a strong emphasis on audio processing and speech synthesis, and was consolidated into a single, targeted feature delivered over the course of one month.