A tab seperated list where each line is like <token> \t <source> \t <phones> where the source is a rudementary identifier for the source of the token and phones is a tab-seperated list of word-phone-predictions, each of which is space seperated. An example:
| Token | Source | Phones |
|---|---|---|
| Kettir verði ekki bundnir í árborg | malromur | cʰ ɛ h t ɪ r <\t> v ɛ r ð ɪ <\t> ɛ h c ɪ <\t> p ʏ n t n ɪ r <\t> iː <\t> au r p ɔ r k |
| Níu grunaðir um ölvunarakstur | malromur | n i j ʏ <\t> k r ʏː n a ð ɪ r <\t> ʏ m <\t> œ l v ʏ n a r a k s t ʏ r |
Using the complete list of valid diphones in data/diphones/complete_ipa.txt, the tokens in all_tokens_norm_clean_g2p.txt cover approx 81% of the diphones. The ivona dataset alone (8909 tokens), approx 10x smaller than the complete list, covers approx 76 % of the listed valid diphones.