Zum Inhalt springen
ModelAufgelistetExtern
Herunterladen

Omnivoice Swahili 1h

OmniVoice (k2-fsa) finetuned for Swahili at a 1h data budget - companion checkpoint for the Deep Learning Indaba 2026 (RIAD) poster "Tone on a Budget".

ArtText modelVersionmodelCBewertung

Modellquelle

Art
Text model
Dateigröße
2,28 GB
Version
vmodel
Aktualisiert
19. Okt. 2026
Downloads
4
Quelle
Hugging Face

Quellenbeschreibung

omnivoice-swahili-1h

OmniVoice (k2-fsa) finetuned for Swahili at a 1h data budget - companion checkpoint for the Deep Learning Indaba 2026 (RIAD) poster "Tone on a Budget".

Code, the reference-free tone_i2 tone metric, and the recipe: https://github.com/mosesdaudu001/tone-on-a-budget

License & codec

These weights are a derivative of OmniVoice (k2-fsa, Apache-2.0). At inference the model loads the Higgs-Audio-v2 codec separately; that codec carries the Boson Higgs Audio 2 Community License (non-commercial, output-restriction). This repo contains only the Apache-2.0-derived OmniVoice weights - obtain the codec from its own source.

Quellen

1 Quelle
Hugging FacePrimärVersion 8fd71774580ec1c16bbddf3d0b60cc2ac18ca924
huggingface.co/mosesdaudu/omnivoice-swahili-1h/tree/8fd71774580ec1c16bbddf3d0b60cc2ac18ca924

Verifiziert 29. Aug.

Modellartefakte

1 Artefakt

model.safetensors

safetensors · 2,28 GB · SHA-256 24991928997e…174f · Hugging Face

Download file

Quellenauszüge

2 Auszüge

--- license: apache-2.0 library_name: omnivoice language: - sw tags: - text-to-speech - tone - low-resource - african-languages - swahili --- # omnivoice-swahili-1h OmniVoice (k2-fsa) finetuned for **Swahili** at a **1h** data budget - companion checkpoint for the Deep Learning Indaba 2026 (RIAD) poster *"Tone on a Budget"*. Code, the reference-free `tone_i2` tone metric, and the recipe: https://github.com/mosesdaudu001/tone-on-a-budget ## License & codec These weights are a derivative of **OmniVoice** (k2-fsa, Apache-2.0). At inference the model loads the **Higgs-Audio-v2 codec** separately; that codec carries the **Boson Higgs Audio 2 Community License** (non-commercial, output-restriction). This repo contains only the Apache-2.0-derived OmniVoice weights - obtain the codec from its own source.

Source context: 4 downloads · 0 likes · Pipeline text-to-speech · Library omnivoice · Repo mosesdaudu/omnivoice-swahili-1h