CraBERT: A Cascade Pre-Trained Phoneme Encoder with Subword Representations Alignment for Text-to-Speech



For each pair, the first row corresponds to Table I and the second row corresponds to Table II from our paper.
Notably, the 75% in Table I is also the CraBERT-1e in Table II.

For convenience, we display only the subword sequence, where ## is the subword prefix, / is the respiratory pause mark, and ^ is the breath mark.

1.(speaker id: 278) ill / as he had fare ##d in his interview with the archdeacon ^ he was not the less under the necessity of carrying out his pledge to / eleanor
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
2.(speaker id: 688) ordinary phosphorus is soluble in oil , alcohol , ether , ch ##lor ##of ##or ##m , and carbon di ##sul ##phi ##de ; ins ##ol ##ub ##le in water ^ it is much used in rat poison ##s
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
3.(speaker id: 1018) that reminds me of ^ of something i wanted to say to you , she said ^ you re quite right about people not rep ##ress ##ing their inclination ##s
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
4.(speaker id: 1212) first he released the little dog ' s head ^ which had been bent down between his fore legs
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
5.(speaker id: 1779) the whole meta ##physics of / aristotle ^ is dominated by the / distinction between form ^ and matter
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
6.(speaker id: 3307) in order to discover if the thom ##is ##us is capable ^ of a similar error ^ i gathered some broken pieces of silk worm s coco ##on into a closed / cone ^ turning the fragments /
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
7.(speaker id: 3549) but when they came / to make the experiment , they were disappointed of their hope ^ for the miners could make but small progress ^ and that with difficulty also
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
8.(speaker id: 3945) philip had to walk some five miles before he reached a little station , where he could wait for a train , and he had ample time for reflection
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
9.(speaker id: 3977) ^ and ph ##logist ##icated air less ; but after passing through the water ^ they appeared / not to have been sen ##si ##bly changed by the process
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
10.(speaker id: 4290) joy appeared to have passed away as a factor of life ^ and this creative impulse had nothing to take its place
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
11.(speaker id: 4290) he was a very keen person and i m afraid ^ he chuckled with / evident satisfaction to himself ^ i m really afraid , mr glen ##arm that you re not
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline
12.(speaker id: 5189) there s no law in christ ##end ##om against bas ##ting a man with a roast pig ! dick s weapon failed before his anger
15% 30% 45% 60% 75% (CraBERT-1e) 90%
Ground truth CraBERT-10e CraBERT-0e CraBERTpara-0e MP BERT PL BERT Baseline