godzilla1215/dsn2026-lid-d7

Spoken language identification for 12 Nigerian languages (DSN 2026)

Submission of team "godzilla" to the DSN 2026 Find the Language I Am Speaking competition.

A standard HubertForSequenceClassification model (loads with AutoModelForAudioClassification, no custom code). It predicts one of 13 labels: yor ibo hau ego anw efi ibb idu igl nup tiv urh other.

Approach

AfriHuBERT (all 12 transformer layers) with a weighted layer sum and mean pooling, fine-tuned for 8,000 steps on random 3-6 s crops with speaker- and source-balanced sampling (external sources capped at half of a class) and audio augmentation (synthetic noise, reverberation, telephone band-limiting and companding, speed and equalisation changes). Training labels were cleaned (clearly English segments relabelled as other). Per-class decision offsets, tuned on speaker-held-out cross-validation and then corrected using the public-leaderboard confusion matrix of the previous submission, are added to the classifier bias. The released weights are the average of two such models trained from the same initialisation with different data order and augmentation seeds (one standard model of the same size). Inference uses the full clip. Training data additionally includes Common Voice 27.0 Ibibio and the Mozilla Data Collective Nupe speech collection, after a speaker-overlap check against the competition data.

Base model

• ajesujoba/AfriHuBERT, a continued pretraining of utter-project/mHuBERT-147 (CC BY-NC-SA 4.0). This model inherits that licence.

Training data

• Competition training and validation data, Data-Science-Nigeria/dsn-2026-lid-train, which combines afrispeech_ipa (CC-BY-4.0), hypa_voices (Apache-2.0) and omnilingual_asr (CC-BY-4.0); see its ATTRIBUTION.md. No other use of those corpora.

External datasets (openly licensed; per-dataset licence shown):

• offiongbassey/efik_speech_dataset, default/train, revision f7e84a96ee, CC-BY-4.0, used as efi
• google/fleurs, ar_eg/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, en_us/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, es_419/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, ff_sn/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, fr_fr/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, ha_ng/train, revision 70bb2e84b9, CC-BY-4.0, used as hau
• google/fleurs, ig_ng/train, revision 70bb2e84b9, CC-BY-4.0, used as ibo
• google/fleurs, lg_ug/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, ln_cd/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, pt_br/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, sw_ke/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, wo_sn/train, revision 70bb2e84b9, CC-BY-4.0, used as other
• google/fleurs, yo_ng/train, revision 70bb2e84b9, CC-BY-4.0, used as yor
• McGill-NLP/NaijaS2ST, default/dev+train language=english, revision 4fef5310ef, CC-BY-4.0, used as other
• McGill-NLP/NaijaS2ST, default/train language=pidgin, revision 4fef5310ef, CC-BY-4.0, used as other
• asr-nigerian-pidgin/nigerian-pidgin-1.0, default/train, revision 87c6150918, CC-BY-4.0, used as other
• google/WaxalNLP, aka_asr/train, revision 5f4d8ca24f, CC-BY-4.0 (University of Ghana), used as other
• google/WaxalNLP, bau_tts/train, revision 5f4d8ca24f, CC-BY-4.0 (University of Ghana), used as other
• google/WaxalNLP, dag_asr/train, revision 5f4d8ca24f, CC-BY-4.0 (University of Ghana), used as other
• google/WaxalNLP, ewe_asr/train, revision 5f4d8ca24f, CC-BY-4.0 (University of Ghana), used as other
• google/WaxalNLP, ful_asr/train, revision 5f4d8ca24f, CC-BY-SA-4.0 (Digital Umuganda), used as other
• google/WaxalNLP, pcm_tts/train, revision 5f4d8ca24f, CC-BY-SA-4.0 (Media Trust), used as other
• naijavoices/voiceafrica-datasets, igl/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as igl
• naijavoices/voiceafrica-datasets, nup/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as nup
• naijavoices/voiceafrica-datasets, urh/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as urh
• naijavoices/voiceafrica-datasets, knc/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, bcz/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, ble/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, bvb/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, fan/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, krx/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, pov/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• naijavoices/voiceafrica-datasets, srr/train, revision 8e1e831d00, CC-BY-NC-SA-4.0 + NaijaVoices terms (https://naijavoices.com/terms-and-conditions); no voice cloning, used as other
• Wikimedia Commons (Wikimedia Nigeria oral-history project): 44 recordings (CC BY 4.0, CC BY-SA 4.0); speech segments only, English and music removed:
  • The audiovisual documentary of the Ibibio (Ibesikpo Asutan) dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as ibb
  • The audiovisual documentary of the Ibibio (Ini) dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as ibb
  • The audiovisual documentary of the Ibibio (Nsit Ubium) dialect by a native speakers.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as ibb
  • The audiovisual documentary of the Ibibio live dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as ibb
  • The audiovisual documentary of Annang dialect by a Native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as anw
  • The audiovisual documentary of the Annang (Ika) dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as anw
  • The audiovisual documentary of Efik dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as efi
  • The audiovisual documentary of Efik language by its native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as efi
  • The audiovisual documentary of the Ughelli dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Uwheru dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Orogun dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Agbarha dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Agbon dialect by speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Evwreni dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Idjerhe dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Effurun-Otor dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Ughievwen dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Olomu dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Okparabe dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of the Okere dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as urh
  • The audiovisual documentary of Oro dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Uruefong dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Idua dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Usem Okobuo dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Eket dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Esit Eket by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Ibeno dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Egbughu dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Obolo language by its native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Itshekiri dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of the Izon dialect by native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • A short oral history of Degema in Degema language by a native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Engenni language by its native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Kalabari language by its native speaker..webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Nembe language by native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Okrika language by its native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Khana language by its native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Ogoni language in Gokana Bodo dialect by its native speaker.webm, CC BY 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • A short oral history of Abua in Emughan-Abua dialect by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Bekwarra language by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Boki-west language by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Ejagam Bendege by a native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Nbembe language by its native speaker.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
  • The audiovisual documentary of Yala language by its native speakers.webm, CC BY-SA 4.0, by Wikimedia Nigeria Foundation Inc, used as other
• Wikitongues (via Wikimedia Commons): 4 recordings (CC BY 4.0, CC BY-SA 4.0, CC0); speech segments only, English and music removed:
  • Itoro speaking Ibibio.webm, CC BY-SA 4.0, by Wikitongues Inc, used as ibb
  • Tochi speaking Nigerian Pidgin.webm, CC0, by Wikitongues, Inc, used as other
  • Umar speaking Central Kanuri.webm, CC BY 4.0, by Wikitongues, Inc, used as other
  • Abdulrahman speaking Baatonum.webm, CC BY-SA 4.0, by 2025-5-5, used as other
• Bloom Library talking books: 2 recordings (CC BY 4.0); speech segments only, English and music removed:
  • Bloom: Ori O̱há Atrà (Eggon talking book), CC BY 4.0, by 2024, SIL Nigeria, used as ego
  • Bloom: Kwav u Maria (Tiv talking book), CC BY 4.0, by 2024, Emmanuel Orakaaga, used as tiv
• Common Voice cv-corpus-27.0-2026-09-17 (ibb), validated, revision 27.0-2026-, CC0-1.0, used as ibb
• Speech Data Collection for The Nupe Language (Mozilla Data Collective, 7 speakers), all, revision mdc, CC-BY-NC-SA-4.0, used as nup

Usage

from transformers import AutoFeatureExtractor, AutoModelForAudioClassification
fe = AutoFeatureExtractor.from_pretrained('godzilla1215/dsn2026-lid-d7')
model = AutoModelForAudioClassification.from_pretrained('godzilla1215/dsn2026-lid-d7')
inputs = fe(waveform_16khz_mono, sampling_rate=16000, return_tensors="pt")
label = model.config.id2label[model(**inputs).logits.argmax(-1).item()]