Language material for English audiovisual speech recognition system development

• DOI: 10.1121/1.4830856
• Publication Date: 2013-11-01T23:32:03Z
• Authors: Andrzej Czyzewski; Tomasz Ciszewski; Dorota Majewicz; Bozena Kostek
• Keywords: Sample (material)

The bi-modal speech recognition system requires a 2-sample language input for training and testing algorithms which precisely depicts natural English speech. For the purposes of audio-visual recordings, data base 264 sentences (1730 words without repetitions; 5685 sounds) has been created. sample reflects vowel consonant frequencies in recording material both lexical word casual sound BNC corpus approx. 100m words. semantically syntactically congruent mirror 100m-word frequencies. absolute deviation from source is.09% individual is reduced to level between 0.0006% (min.) 0.009% (max.). 0.006% oscillates 0.00002% 0.012% Similar convergence achieved (29 sentences; 599 sounds). post-recording analysis involves examination particular articulatory settings aid visual as well co-articulatory processes may affect acoustic characteristics sounds. Results elements employing are included paper.