Adding video-derived visual features to FastSpeech2 improves pitch, energy, and duration prediction on a 33-hour movie TTS dataset, with UTMOS rising from 2.91 to 3.13.
Experimental Setup To assess the impact of visual information on speech generation performance, a dataset containing both diverse prosodic and vi- sual data is essential
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
VisualSpeech: Enhancing Prosody Modeling in TTS Using Video
Adding video-derived visual features to FastSpeech2 improves pitch, energy, and duration prediction on a 33-hour movie TTS dataset, with UTMOS rising from 2.91 to 3.13.