Webscrape text for making a word list and a language model suitable for Kaldi ASR.
To collect about 60 files named wikipedia/*/yyyymmdd.txt, run crawl_wikipedia_all_lang.
Filter more, to be appropriate for pseudo-swahili ASR:
- Replace numbers with newlines.
- Cull lines with fewer than 3 words?
Scrape more than just the top page?