A factorized speech codec with separate acoustic, phonetic, and lexical token streams shows that a lexical token layer can act as a word-level detector while preserving usable reconstruction quality.
Like Encodec, ST updates its codebooks via exponential moving average (EMA) and period- ically re-initializes them to maximize utilization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Factorized RVQ-GAN For Disentangled Speech Tokenization
A factorized speech codec with separate acoustic, phonetic, and lexical token streams shows that a lexical token layer can act as a word-level detector while preserving usable reconstruction quality.