w2v-bert-2.0-lang-id-lug-sna-lin-xog

A spoken language identification (LID) model that classifies short audio clips into one of four languages — Luganda, Shona, Lingala, or Lusoga — fine-tuned from facebook/w2v-bert-2.0 with a sequence classification head, trained on the WAXAL corpus only. This extends the 3-language keystats/w2v-bert-2.0-lang-id-lug-sna-lin family with Lusoga added as a fourth class.

Model description

facebook/w2v-bert-2.0 — a large-scale, multilingual self-supervised speech encoder pretrained with a BERT-style masked prediction objective — is used as the backbone, with a from-scratch audio classification head fine-tuned to predict which of the four languages a given audio clip is spoken in. Like the other LID models in this collection, this model does not transcribe speech — it only outputs a language label.

Training data

Only WAXAL data was used — no additional community datasets were pooled in, unlike the -best LID variant for the 3-language set.

Training procedure

• Base model: facebook/w2v-bert-2.0
• Architecture: Wav2Vec2BertForSequenceClassification (audio classification head over the pretrained encoder), 4 output classes (Luganda, Shona, Lingala, Lusoga)
• Sample rate: 16 kHz mono
• Hardware: single RTX PRO 6000

Evaluation results

This model scored ~99.9% accuracy on both the WAXAL validation and test splits, with only a small handful of clips misclassified out of the full evaluation sets.

Note: adding a fourth, more closely related language (Lusoga sits in the same broader Bantu family as Luganda, and the two are more acoustically similar to each other than to Shona or Lingala) makes the classification task somewhat harder than the 3-language case, which likely explains the small gap versus the 3-language model's 100% on WAXAL. As always, treat in-domain WAXAL numbers as an upper bound rather than a guarantee of real-world performance on out-of-domain audio.

How to use

import torch
import librosa
from transformers import Wav2Vec2BertForSequenceClassification, Wav2Vec2BertProcessor

MODEL_ID = "keystats/w2v-bert-2.0-lang-id-lug-sna-lin-xog"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

processor = Wav2Vec2BertProcessor.from_pretrained(MODEL_ID)
model = Wav2Vec2BertForSequenceClassification.from_pretrained(MODEL_ID).to(DEVICE).eval()

audio_array, sr = librosa.load("path/to/audio.wav", sr=16000, mono=True)
inputs = processor(audio_array, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
    logits = model(input_features=inputs.input_features.to(DEVICE)).logits

predicted_id = torch.argmax(logits, dim=-1).item()
predicted_label = model.config.id2label[predicted_id]

print(predicted_label)  # "luganda", "shona", "lingala", or "lusoga"

Intended uses & limitations

• Intended for identifying which of Luganda, Shona, Lingala, or Lusoga is being spoken in a given audio clip — useful as a pre-processing step to route audio to the correct language-specific ASR model (e.g. the Luganda -main-best or Lusoga -main-best checkpoints in this collection).
• This is a closed-set 4-way classifier: it will always output one of the four trained languages, even for audio in a different language or non-speech audio. It has no "none of the above" / out-of-domain detection.
• Trained and evaluated only on WAXAL data — near-perfect validation/test accuracy reflects in-domain WAXAL audio conditions and may not carry over as-is to audio from very different recording conditions, speakers, or dialectal variation.
• Not a general-purpose language ID model — it only distinguishes between these four specific languages, not the full range of African languages or languages generally.

Citation

If you use this model, please cite the training/fine-tuning work and the underlying dataset:

@misc{keystats_wav2vec2bert_langid_lug_sna_lin_xog,
  title={w2v-bert-2.0-lang-id-lug-sna-lin-xog: A Luganda/Shona/Lingala/Lusoga spoken language identification model fine-tuned from facebook/w2v-bert-2.0},
  author={keystats},
  year={2026},
  howpublished={\url{https://huggingface.co/keystats/w2v-bert-2.0-lang-id-lug-sna-lin-xog}}
}

@misc{waxal,
  title={WAXAL: A Multilingual African Speech Dataset},
  author={Google},
  howpublished={\url{https://huggingface.co/datasets/google/WaxalNLP}}
}

@inproceedings{w2vbert2,
  title={Seamless: Multilingual Expressive and Streaming Speech Translation},
  author={Seamless Communication and others},
  year={2023},
  howpublished={\url{https://huggingface.co/facebook/w2v-bert-2.0}}
}