> Authors: Sreeram Sridhar, Adithya L Bhat, Sasank Chilamkurthy We wanted to fine-tune a Kannada TTS model using openly available data and open tools. Here’s exactly how we did it from Wikipedia dump to a trained model. Code to run this in syntheticdata` folder. We document the process here. — Go to the Wikipedia database download page: https://en.wikipedia.org/wiki/Wikipedia:Database_download — Navigate to the dump site:https://dumps.wikimedia.org/ — For a list of all available dumps: https://dumps.wikimedia.org/backup-index.html — For Kannada wikipedia (example date 2025-08-01): https://dumps.wikimedia.org/knwiki/20250801/ — Download the .bz2 file, extract it to get XML files containing the raw wiki content. We used this GitHub repo to convert Wikipedia XML to plain text: https://github.com/daveshap/PlainTextWikipedia This tool parses XML and removes wiki markup to produce readable plain text in JSON format. We used the Indic NLP library to split the large text corpus into clean Kannada sentences. Indic NLP Library (tokenizer): https://github.com/anoopkunchukuttan/indicnlplibrary — Use the Indic NLP sentence tokenizer to break paragraphs into sentences. It handles…
Модальности:
Генерация текста
Задача: Генерация текста
Автор: chsasank
Теги: parler_tts, kn, endpoints_compatible
Лайков: 4 | Загрузок: 9
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
