Replace R16 tokenizer with final LS+Emilia checkpoint (epoch_22)

README.md

@@ -19,3 +19,14 @@ audio -> `streaming_asr` (word chunks + TextGrids) + `boundary_classifier`
 
 Only `rvq_teacher` is R16-specific; the ASR / alignment / boundary
 components are shared across R8/R16/R32.
+
+## Tokenizer checkpoint
+
+`rvq_teacher/epoch_22.pt` is the final R16 tokenizer, trained on LibriSpeech
+plus the full Emilia set with a learnable codebook at a 10x codebook learning
+rate (`260512_RVQ_R16_C512_nodistill_codelr10x_LS_emilia_full_from_nofsq_e13`).
+It replaces the earlier LibriSpeech-only `epoch_16.pt`.
+
+Load it with `RVQ_R=16` and `RVQ_CODEBOOK_SIZE=512`, and leave
+`RVQ_CODEBOOK_DIM` unset: `codebook_dim` defaults to `feat_dim=256`, which is
+what these weights expect.

rvq_teacher/{epoch_16.pt → epoch_22.pt}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d19d648a8f7ee889fca2de33dcc19212aad2f235b86fe55e39c194dc8cfc89a1
+oid sha256:9dc8f478b05a730817a4d562be7f54871b7abfd624ac9cb1d3bfbf69d1c7bd6b
 size 946776221