ANVE: the private AI office, on the record. See what that means

ANVE

Offline AI in your language · updated 2026-10-16

Can offline AI handle Dogri?
डोगरी in speech, text and translation: missing from global models, covered by AI4Bharat’s.

Short answer. Narrowly, through AI4Bharat. Dogri is not in Whisper, NLLB-200 or Qwen3’s list, and Meta’s MMS coverage table does not list it, although the IndicVoices paper reports an MMS result of 77.6% for it. AI4Bharat’s IndicConformer lists Dogri, and IndicASR scored 33.3% word error rate on natural Dogri speech in that paper. Translation is the stronger side: IndicTrans2 scored 53.9 chrF++ from English to Dogri on conversational sentences, against 40.1 for Google Translate in the same 2023 paper. ANVE has not tested Dogri yet. Coverage and figures are as of October 2026.

01

What open models cover in Dogri

TaskModelsWhat the source says
Speech to text: Whisper, MMSWhisper, Meta MMSDogri is not among Whisper’s 100 languages. Meta’s MMS coverage table does not list Dogri either, although the IndicVoices paper reports an MMS result for it, so treat MMS coverage as unclear.
Speech to text: Indian-language modelsAI4Bharat IndicConformerIndicConformer lists Dogri (doi; MIT licence). IndicVoices, natural speech: IndicASR 33.3%, trained on 65 hours; MMS 77.6%; no Whisper figure. IndicWhisper leaves Dogri out of its 12 languages.
Chat, summaries, draftingSarvam, IndicBERT-v3, Qwen3, LlamaQwen3’s list has no Dogri, and the Llama 3.3 and 4 cards, the gpt-oss card and MILU cover none. Sarvam 30B and 105B list Dogri (doi) in their card metadata, and AI4Bharat’s IndicBERT-v3 adds it in a low-resource training phase; that model is an encoder for classification and search, not a chat model.
TranslationIndicTrans2, Sarvam-Translate, IndicTrans3-betaIndicTrans2 paper (2023), conversational set (IN22-Conv), chrF++: English to Dogri 53.9, Dogri to English 65.0; Google Translate scored 40.1 and 62.9. On web and Wikipedia sentences (IN22-Gen): 57.8 and 72.6 against 47.8 and 67.3. No other open system in the paper’s tables supports Dogri. Sarvam-Translate lists Dogri, and IndicTrans3-beta gives it preliminary support.
02

Devanagari, Jammu and a small Census count

Dogri (डोगरी) is written in Devanagari in the sources above (doi_Deva). The 2011 Census counted 25,96,767 people with Dogri as their mother tongue, 0.21% of India and the eighteenth largest group, up 13.76% from 2001. Jammu and Kashmir had 25,13,712 of them. Himachal Pradesh had 23,609, Punjab 18,155, Delhi 6,350 and Rajasthan 5,120. The Home Ministry says Dogri joined the Eighth Schedule in 2004. The 1991 Census was not held in Jammu and Kashmir, so there is no Dogri figure for that year.

Dogri and Hindi share a script, so a reader has to mark any transcript word that a Hindi-trained model has substituted. None of the sources tests that.

For English meetings, the Svarah paper found Whisper large at 8.0% word error rate on speakers whose mother tongue is Dogri, against 5.3% for native Hindi speakers.

03

What to test before you rely on it

  • Start with IndicConformer for speech and IndicTrans2 for translation; there is little else to compare them with.
  • Record natural Dogri, not read sentences, from speakers in Jammu and from outside it.
  • Check the transcript for Hindi words that replaced Dogri ones, and for spellings your documents do not use.
  • Translate a notice in both directions and ask a Dogri reader to mark errors; the paper’s scores are overlap measures, not human ratings.
  • Treat chat in Dogri as unverified: the cards checked report no benchmark for it.
04

Where ANVE stands

ANVE’s meeting-minutes maker handles Hindi, English and Hinglish today. Dogri has not been tested on ANVE yet, so there is no ANVE result to report and nothing here promises support.

A cloud model is optional on ANVE: it runs only after an administrator adds a provider, its answers are labelled, and each call is logged as local or cloud. A policy that blocks confidential data from cloud models is not on ANVE yet, so until it ships treat any cloud route for Dogri as a conscious exception.

A demo call is free. If Dogri decides your purchase, ask for your own recordings to be tried before you commit.

05

Questions people ask

Does Whisper support Dogri?
No. Dogri is not in Whisper’s list of 100 languages, and neither is it in FLORES-200, which defines NLLB-200’s coverage. AI4Bharat’s IndicConformer and IndicTrans2 list it.
Is there an offline Dogri speech model?
AI4Bharat’s IndicConformer lists Dogri (MIT licence). In the IndicVoices paper IndicASR scored 33.3% on natural Dogri speech, trained on 65 hours.
Can offline AI translate Dogri?
Yes. IndicTrans2 scored 53.9 chrF++ from English to Dogri and 65.0 from Dogri to English on conversational sentences, against 40.1 and 62.9 for Google Translate in that 2023 paper. Have a Dogri reader check anything official.
Do Qwen or Llama models know Dogri?
Neither lists it: Qwen3’s launch post and Meta’s Llama 3.3 and 4 cards omit Dogri. Sarvam’s 30B and 105B card metadata lists it, but no benchmark figure for Dogri chat appears in the cards checked.
06

What is not real yet

No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.

Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.

07

Sources