Table 3 Comparison between the SpeechDat telephone quality (TF), SpeeCon narrow-band (NB) and SpeeCon wide-band (WB) recognisers. Results are given in terms of correct frames % for phonemes (ph) and visemes (vi), and accuracy.
From: SynFace—Speech-Driven Facial Animation for Virtual Speech-Reading Support
Database | SpeechDat | SpeeCon | |
|---|---|---|---|
Data size (ca. hours) | 200 | 40 | |
Speakers (#) | 5000 | 550 | |
Speech quality | TF | NB | WB |
Sampling (kHz) | 8 | 8 | 16 |
Correct frames (% ph) | 54.2 | 65.2 | 68.7 |
Correct frames (% vi) | 59.3 | 69.0 | 74.5 |
Accuracy (% ph) | 56.5 | 62.2 | 63.2 |