eduhk-compling/11609239_Sajana1160

Dataset Description

This dataset is a small Nepali text‑to‑speech corpus. It contains segmented audio with aligned transcriptions for TTS training. The sentences are an original short narrative in Nepali describing a quiet village morning walk, including scenes of the sky, people going to work, children playing, and natural sounds like birds and water. Each sentence has a corresponding audio recording and a romanized transcription, organized in (metadata.csv) following the specified (file_name,hon,rom format).

Issues Encountered and Solution

Organising all the sentences and their English transcriptions into the required file format and ensuring that there is no extra comma was notably time‑consuming. I had to ensure every audio file name, Nepali sentences, and romanization aligned wholly, with consistent use of delimiters and no typos. Double‑checking that the text matched the actual recordings added another coating of work. It also took extensive amount of time to fully understand and recall each step in the workflow, from sectioning decisions to how exactly the metadata line should look. Overall, the process required thorough attention and several rounds of revision before everything felt appropriate.