Counterfactual gradient edits to a pretrained TTS model's encoder activations can control prosody and correct mispronunciations at inference time, at least on Tacotron 2.
What would the internal representation look like if the synthesized speech had a higher pitch rather than a lower pitch?
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Counterfactual Activation Editing for Post-hoc Prosody and Mispronunciation Correction in TTS Models
Counterfactual gradient edits to a pretrained TTS model's encoder activations can control prosody and correct mispronunciations at inference time, at least on Tacotron 2.