LILAC is a fully convolutional 24 kHz speech codec whose invertible analysis transform and finite scalar quantization make every decode-re-encode cycle return the same token stream by construction.
High-Fidelity Audio Compression with Improved
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SD 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LILAC: An Idempotent Neural Speech Codec
LILAC is a fully convolutional 24 kHz speech codec whose invertible analysis transform and finite scalar quantization make every decode-re-encode cycle return the same token stream by construction.