The new Facebook AI has found a way to detect phonemes without prior samples. On the TIMIT benchmark, Wav2vec-U reduced the average error rate from 26.1 to 11.3 in terms of the percentage of phonemes that are guessed right. Wav2ves-U, when tested with “low resource” languages like Tatar, Amharic, Swahili, Kyrgyz, outperformed some of the benchmark supervised-learning programs that used labeled data.