DSN 2026 spoken language ID (Whisper-base encoder)

Spoken language identification for 12 Nigerian languages plus an other class, for the DSN 2026 "Find the Language I Am Speaking" competition.

Variant

• Same weights as the calibrated model, with the per-class calibration offsets removed.

Approach

• Encoder of openai/whisper-base, fine-tuned end to end with WhisperForAudioClassification (weighted layer sum over encoder layers).
• Input window of 10 s: the processor pads or truncates every clip to its first 10 s, and the encoder keeps the matching first positions of Whisper's position table.
• Training: rare-language oversampling (inverse-sqrt class frequency), random crops, waveform augmentation (speed, gain, reverberation, telephone band-limiting with mu-law, coloured noise, babble from other training clips), SpecAugment, label smoothing 0.1.
• Per-class logit offsets tuned to maximise language-weighted F1, kept only if they also improve held-out halves of the validation split, and folded into the classifier bias.

Training data

• Train and validation splits combined, for the epoch count chosen by the train-only run. Calibration offsets were reused from that run.
• Source: Data-Science-Nigeria/dsn-2026-lid-train. Its sources keep their own licences; see the dataset's ATTRIBUTION.md and LICENSE.md.

External data

• None. No prohibited source was used.

Validation (language-weighted F1)

• note: trained on train + validation, so it has no held-out score
• epochs: 7
• main run val language-weighted F1 (calibrated): 0.6811

Limitations

• Nupe has no validation clips, so its calibration offset is left at zero.
• Training data is mostly read speech, including religious recordings, so conversational speech may score lower.
• Licence of this repo: MIT, which is the licence of the base Whisper model. The training data's licences are separate.