From: Advanced recurrent network-based hybrid acoustic models for low resource speech recognition
Language | # words | # phonemes | LM perplexity |
---|---|---|---|
Cantonese | 18512 | 216 | 124.87 |
Pashto | 17646 | 126 | 217.55 |
Vietnamese | 6210 | 375 | 176.99 |
Swahili | 21890 | 75 | 435.74 |
Tamil | 52369 | 34 | 656.51 |
Kazakh | 19587 | 133 | 476.04 |
Georgian | 34946 | 35 | 471.03 |