From the model card
Fields below are copied from the tags and counters on the HuggingFace repository k2-fsa/OmniVoice at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- k2-fsa
- Pipeline tag
- text-to-speech
- Framework tags
- JAX
- Weight formats
- safetensors
- Lineage
-
- base model Qwen/Qwen3-0.6B
- fine-tune of Qwen/Qwen3-0.6B
- Language tags
- multilingual; Abkhazian (ab), Adyghe (ady), Tunisian Arabic (aeb), Afrikaans (af), Judeo-Arabic (aju), Gheg Albanian (aln), Amharic (am), Aragonese (an), Angika (anp), Arabic (arb), Algerian Arabic (arq), Najdi Arabic (ars), Moroccan Arabic (ary), Egyptian Arabic (arz), Assamese (as), Asturian (ast), Azerbaijani (az), Bashkir (ba), Basaa (bas), Bamun (bax), Ghomala (bbj), Belarusian (be), Betawi (bew), Bafut (bfd), Bulgarian (bg), Bhojpuri (bho), Banjar (bjn), Kom (bkm), Bangla (bn), Tibetan (bo), Breton (br), Braj (bra), Brahui (brh), Bodo (brx), Bosnian (bs), Buginese (bug), Bulu (bum), Medumba (byv), Catalan (ca), Cebuano (ceb), Chiga (cgg), Central Kurdish (ckb), Czech (cs), Chuvash (cv), Welsh (cy), Danish (da), Dargwa (dar), Taita (dav), German (de), Dogri (dgo), Zarma (dje), Duala (dua), Divehi (dv), Dyula (dyu), Dazaga (dzg), Embu (ebu), Greek (el), English (en), Esperanto (eo), Spanish (es), Central Yupik (esu), Estonian (et), Basque (eu), Ewondo (ewo), Extremaduran (ext), Persian (fa), Fang (fan), Akan (fat), Fula (ff), Finnish (fi), Filipino (fil), French (fr), Fula (fuc), Western Frisian (fy), Irish (ga), Galician (gl), Guarani (gn), Konkani (gom), Gujarati (gu), Frafra (gur), Gusii (guz), Manx (gv), Gbaya (gya), Hausa (ha), Hawaiian (haw), Hebrew (he), Hindi (hi), Croatian (hr), Upper Sorbian (hsb), Haitian Creole (ht), Hungarian (hu), Armenian (hy), Herero (hz), Interlingua (ia), Ibibio (ibb), Indonesian (id), Igbo (ig), Inupiaq (ik), Icelandic (is), Italian (it), Japanese (ja), Ngomba (jgo), Javanese (jv), Georgian (ka), Kabyle (kab), Jju (kaj), Kamba (kam), Kabardian (kbd), Kanembu (kbl), Kabuverdianu (kea), Khowar (khw), Kuanyama (kj), Kazakh (kk), Kalenjin (kln), Khmer (km), Kurdish (kmr), Kannada (kn), Korean (ko), Kashmiri (ks), Bafia (ksf), Cornish (kw), Kyrgyz (ky), Langi (lag), Luxembourgish (lb), Ganda (lg), Ligurian (lij), Lingala (ln), Lao (lo), Lithuanian (lt), Latgalian (ltg), Luba-Lulua (lua), Luo (luo), Mizo (lus), Latvian (lv), Mafa (maf), Maithili (mai), Maba (mde), Moksha (mdf), Meru (mer), Mari (mhr), Māori (mi), Macedonian (mk), Malayalam (ml), Mongolian (mn), Manipuri (mni), Marathi (mr), Western Mari (mrj), Malay (ms), Maltese (mt), Mundang (mua), Burmese (my), Erzya (myv), Min Nan Chinese (nan), Neapolitan (nap), Norwegian Bokmål (nb), Ndonga (ng), Dutch (nl), Kwasio (nmg), Norwegian Nynorsk (nn), Ngiemboon (nnh), Norwegian (no), Nepali (npi), Northern Sotho (nso), Nyanja (ny), Occitan (oc), Oromo (om), Odia (ory), Ossetic (os), Punjabi (pa), Pashto (pbu), Nigerian Pidgin (pcm), Polish (pl), Malagasy (plt), Piedmontese (pms), Western Panjabi (pnb), Pashto (ps), Portuguese (pt), Chimborazo Highland Quichua (qug), Romansh (rm), Romanian (ro), Rombo (rof), Russian (ru), Aromanian (rup), Kinyarwanda (rw), Sanskrit (sa), Yakut (sah), Santali (sat), Sardinian (sc), Sicilian (scn), Sindhi (sd), Seri (sei), Chadian Arabic (shu), Sinhala (si), Slovak (sk), Slovenian (sl), Shona (sn), Soninke (snk), Somali (so), Albanian (sq), Serbian (sr), Sardinian (src), Swedish (sv), Swahili (sw), Tamil (ta), Tulu (tcy), Telugu (te), Tajik (tg), Thai (th), Tigrinya (ti), Tigre (tig), Turkmen (tk), Tlingit (tli), Tswana (tn), Toki Pona (tok), Turkish (tr), Taroko (trv), Torwali (trw), Tatar (tt), Akan (tw), Uyghur (ug), Ukrainian (uk), Umbundu (umb), Urdu (ur), Uzbek (uz), Uzbek (uzn), Vai (vai), Vietnamese (vi), Votic (vot), Võro (vro), Wolof (wo), Xhosa (xh), Mingrelian (xmf), Kpelle (xpe), Yangben (yav), Yiddish (ydd), Yiddish (yi), Yoruba (yo), Cantonese (yue), Standard Moroccan Tamazight (zgh), Chinese (zh), Zulu (zu), Zaza (zza)
- Papers cited
- arXiv:2604.00688
- Downloads (HF counter at last fetch)
- 1,013,539
- Likes (HF counter at last fetch)
- 1,334
- Model card
- https://huggingface.co/k2-fsa/OmniVoice
Use cases
- Multilingual ASR in sherpa-onnx based applications
- Speech recognition for embedded and edge deployment
- Integration into k2/icefall training pipelines
- Building multilingual speech interfaces without per-language models
Pros
- Integrates with the sherpa-onnx ecosystem for production deployment
- Multilingual coverage reduces per-language model management
- Compatible with k2 training infrastructure for fine-tuning
- Designed for edge deployment via ONNX runtime
Cons
- Limited documentation outside the k2-fsa ecosystem
- Benchmark comparisons against Whisper or SeamlessM4T not published
- k2 framework has a steep learning curve for newcomers
- License and training data details underdocumented
Tags
omnivoicesafetensorszero-shotmultilingualvoice-cloningvoice-designtext-to-speechaaeaalaaoababbabnabrabsabvacmacwacxadf