Repository navigation
feat(denoise): variante faible latence DeepFilterNet3_ll (opt-in par le modèle) - #279
Merged
Merged
Conversation
… le modèle Le réseau DFN3 est paramétré par une Config (largeur des GRU, couches, noyau temporel interne, groupes, anticipation) lue dans les métadonnées safetensors ; un fichier sans métadonnées reste le DFN3 standard, identique au bit. Les convolutions internes à noyau temporel 2 gardent leurs entrées passées dans l'état du flux. Convertisseur ONNX vers safetensors (poids seuls, graphe jamais exécuté) et oracle onnxruntime.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Résumé
DeepFilterNet3_ll(MIT/Apache) : sans anticipation, retard 10 ms au lieu de 30 (20 ms au lieu de 40 avec le hop en remplissage).network.rsparamétré parConfig(GRU 256/512, couches DF 2/3, noyau temporel interne 1/2, groupesdf_fc_emb32/16, anticipation 2/0), lue dans les métadonnées safetensors ; sans métadonnées : DFN3 standard.tools/voice/convert_dfn3_onnx.py: export ONNX officiel → safetensors (lecture des poids seulement, BN repliée, GRU remis en ordre PyTorch). Sur l'export ONNX standard, il redonne au bit les poids du checkpoint.Vérifications
cmpsurlong.wav(1 et 8 cœurs) contre les sorties de main ; oracle PyTorch inchangé (masque 1,5e-6, 134,6 dB).dfn3_ll_matches_onnxruntime, tolérances figées à 1e-3) : masque 7,8e-7, coefs 2,5e-6, LSNR 2,6e-5.-D warnings, tests lib denoise/enhance/safetensor.🤖 Generated with Claude Code