model card: 2026-08-16 release — UTMOSv2 2.919 (above GT 2.822, level with nemo44k 2.903 at ~1/10 bitrate), CV22 CER unchanged 7.70%; add progression chart

charts: 2026-08-16 release — UTMOSv2 2.788 -> 2.919 on the 9,291-clip wideband set, CV22 CER unchanged at 7.70%

charts: 2026-08-16 release — UTMOSv2 2.788 -> 2.919 on the 9,291-clip wideband set, CV22 CER unchanged at 7.70%

clean-corpus finetune @50k : CER 7.70 macro / 7.09 micro (IDENTICAL to previous weights), wideband UTMOSv2 2.799->2.913 (+0.114), held-out MOS on 138 unseen sources 2.654->2.807 (+0.153, t=14.2), PESQ-WB +0.0170 (t=7.5). Same architecture and recipe; only the training corpus changed (clean DNSMOS-gated >=44kHz, CommonVoice excluded). Previous weights remain in history at sha256 8869a98b...

Update README.md

README: remove GitHub links (code closed-source; inference bundled in repo)

README: self-contained inference quickstart

Add neucodec package for self-contained inference

Add self-contained inference CLI

Update README.md

card: drop PESQ (favors waveform codecs); note multilingual set

plot: annotate UTMOSv2 (drop PESQ)

card: add PESQ column + bitrate-vs-UTMOSv2 plot

add bitrate-vs-UTMOSv2 plot

comparison: drop 24 kHz anchor (>=44k class only)

benchmark: final gs-2.0M WideCodec numbers (UTMOSv2 2.788, beats DAC)

refresh samples/after @ global_step 2000000

full resume ckpt @ global_step 2000000

weights @ global_step 2000000

intro: SOTA 0.8 kbps 44.1 kHz audio tokenizer

add 44k-class bitrate comparison table

full resume ckpt @ global_step 1875000

weights @ global_step 1875000

rename NeuCodec-44k -> WideCodec in model card

remove stray samples/boboiboy/after/ path

fix boboiboy-after path (canonical) @ gs1795000

refresh samples/after @ global_step 1795000

full resume ckpt @ global_step 1785000

weights @ global_step 1785000

README: document stage-1 (noisy) revision + stage-2 latest

full resume ckpt from epoch=24-step=1745000.ckpt

Upload pytorch_model.bin with huggingface_hub

full resume ckpt from epoch=9-step=1580000.ckpt

Upload pytorch_model.bin with huggingface_hub

Upload samples/boboiboy/boboiboy-after.wav with huggingface_hub

samples: after (recon 44.1k d20)

samples: before (originals)

Delete folder samples/common-voice with huggingface_hub

full resume ckpt from epoch=9-step=1565000.ckpt

Upload pytorch_model.bin with huggingface_hub

Upload samples/boboiboy/boboiboy-after.wav with huggingface_hub

samples: after (recon 44.1k d20)

samples: before (originals)

Delete folder samples/common-voice with huggingface_hub

full resume ckpt from epoch=9-step=1510000.ckpt

Upload pytorch_model.bin with huggingface_hub

Upload samples/boboiboy/boboiboy-after.wav with huggingface_hub

samples: after (recon 44.1k d20)

samples: before (originals)

Delete folder samples/common-voice with huggingface_hub