kreasof-ai/whisper-medium-bem2eng

whisper-medium-bem2en

This model is a fine-tuned version of openai/whisper-medium on the Big-C Dataset and Bemba-Speech. It achieves the following results on the evaluation set:

• Loss: 0.6966
• Wer: 38.3922

Model description

This model is a transcription model for Bemba Audio.

Intended uses

This model was used for the Bemba-to-English translation task as part of the IWSLT 2025 Low-Resource Track.

Training and evaluation data

This model was trained using the train+dev split from BembaSpeech Dataset and train+val split from Big-C Dataset. Meanwhile for evaluation, this model used test split from Big-C and BembaSpeech Dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.172	1.0	6205	0.5755	47.5724
0.8696	2.0	12410	0.4932	40.5547
0.6827	3.0	18615	0.4860	38.7776
0.3563	4.0	24820	0.5455	38.3652
0.1066	5.0	31025	0.6966	38.3922

Model Evaluation

Performance of this model was evaluated using WER on the test split of Big-C dataset.

Finetuned/Baseline	WER
Baseline	150.92
Finetuned	36.19

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.4.0
• Tokenizers 0.21.0

Citation

@misc{radford2022whisper,
  doi = {10.48550/ARXIV.2212.04356},
  url = {https://arxiv.org/abs/2212.04356},
  author = {Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
  title = {Robust Speech Recognition via Large-Scale Weak Supervision},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}

@inproceedings{sikasote-etal-2023-big,
    title = "{BIG}-{C}: a Multimodal Multi-Purpose Dataset for {B}emba",
    author = "Sikasote, Claytone  and
      Mukonde, Eunice  and
      Alam, Md Mahfuz Ibn  and
      Anastasopoulos, Antonios",
    editor = "Rogers, Anna  and
      Boyd-Graber, Jordan  and
      Okazaki, Naoaki",
    booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.acl-long.115",
    doi = "10.18653/v1/2023.acl-long.115",
    pages = "2062--2078",
    abstract = "We present BIG-C (Bemba Image Grounded Conversations), a large multimodal dataset for Bemba. While Bemba is the most populous language of Zambia, it exhibits a dearth of resources which render the development of language technologies or language processing research almost impossible. The dataset is comprised of multi-turn dialogues between Bemba speakers based on images, transcribed and translated into English. There are more than 92,000 utterances/sentences, amounting to more than 180 hours of audio data with corresponding transcriptions and English translations. We also provide baselines on speech recognition (ASR), machine translation (MT) and speech translation (ST) tasks, and sketch out other potential future multimodal uses of our dataset. We hope that by making the dataset available to the research community, this work will foster research and encourage collaboration across the language, speech, and vision communities especially for languages outside the {``}traditionally{''} used high-resourced ones. All data and code are publicly available: [\url{https://github.com/csikasote/bigc}](\url{https://github.com/csikasote/bigc}).",
}

@InProceedings{sikasote-anastasopoulos:2022:LREC,
  author    = {Sikasote, Claytone  and  Anastasopoulos, Antonios},
  title     = {BembaSpeech: A Speech Recognition Corpus for the Bemba Language},
  booktitle      = {Proceedings of the Language Resources and Evaluation Conference},
  month          = {June},
  year           = {2022},
  address        = {Marseille, France},
  publisher      = {European Language Resources Association},
  pages     = {7277--7283},
  abstract  = {We present a preprocessed, ready-to-use automatic speech recognition corpus, BembaSpeech, consisting over 24 hours of read speech in the Bemba language, a written but low-resourced language spoken by over 30\% of the population in Zambia. To assess its usefulness for training and testing ASR systems for Bemba, we explored different approaches; supervised pre-training (training from scratch), cross-lingual transfer learning from a monolingual English pre-trained model using DeepSpeech on the portion of the dataset and fine-tuning large scale self-supervised Wav2Vec2.0 based multilingual pre-trained models on the complete BembaSpeech corpus. From our experiments, the 1 billion XLS-R parameter model gives the best results. The model achieves a word error rate (WER) of 32.91\%, results demonstrating that model capacity significantly improves performance and that multilingual pre-trained models transfers cross-lingual acoustic representation better than monolingual pre-trained English model on the BembaSpeech for the Bemba ASR. Lastly, results also show that the corpus can be used for building ASR systems for Bemba language.},
  url       = {https://aclanthology.org/2022.lrec-1.790}
}

Contact

This model was trained by Hazim.

Acknowledgments

Huge thanks to Yasmin Moslem for her supervision, and Habibullah Akbar the founder of Kreasof-AI, for his leadership and support.