README.md

---
language: bo
license: apache-2.0
library_name: peft
tags:
- automatic-speech-recognition
- tibetan
- low-resource
- whisper
- lora
- peft
base_model: openai/whisper-tiny
datasets:
- billingsmoore/nict-tib1
metrics:
- cer
- wer
model-index:
- name: tibetan-asr-nict-tib1-whisper-tiny-lora
  results:
  - task:
      type: automatic-speech-recognition
      name: Automatic Speech Recognition
    dataset:
      type: billingsmoore/nict-tib1
      name: NICT-Tib1 (Lhasa Tibetan)
      split: test
    metrics:
    - type: cer
      value: 0.7587
      name: CER
    - type: wer
      value: 1.0849
      name: BoTok-SWER (Segmented WER)
---

# Whisper Tiny — Tibetan ASR (LoRA)

Fine-tuned **`openai/whisper-tiny`** for automatic speech recognition (ASR) on **Lhasa Tibetan**, released alongside:

> J. Moore, S. Li and P. Lauren, "Evaluating Tibetan ASR With Segmented Word Error Rate: Beyond Character-Level Metrics," in *IEEE Access*, vol. 14, pp. 101790-101805, 2026, doi: 10.1109/ACCESS.2026.3709206.

This repo is one of 14 model checkpoints released with the paper, benchmarking 5 ASR architectures (Whisper Tiny/Base/Small, Wav2Vec 2.0 Base, HuBERT Base) under full fine-tuning, LoRA, and QLoRA (8-bit/4-bit) adaptation. See [billingsmoore/tibetan-asr-nict-tib1-*](https://huggingface.co/billingsmoore) for the full set.

## Model description

Whisper is an encoder–decoder transformer, originally pre-trained by OpenAI on 680,000 hours of weakly-labeled multilingual speech-text data, fine-tuned here for sequence-to-sequence Tibetan transcription. This checkpoint uses **LoRA fine-tuning (rank-decomposition adapters, base weights frozen and unquantized)**.

**LoRA configuration:** r=8, alpha=16, dropout=0.1, targeting the attention projections (`k_proj`, `q_proj`, `v_proj`, `out_proj`) and both feed-forward layers (`fc1`, `fc2`) of every Whisper transformer block.

## Training data

Fine-tuned on [NICT-Tib1](https://huggingface.co/datasets/billingsmoore/nict-tib1), a corpus of transcribed Lhasa Tibetan speech from 20 speakers (Soky, Gong & Li, 2022). The data was split by speaker (85/15, not by utterance) so that no speaker appears in both splits: 15,099 training utterances from 17 speakers, 1,547 test utterances from the remaining 3 speakers.

## Training procedure

- Base model: `openai/whisper-tiny`
- Learning rate: 3.75e-5
- Max steps: 4000 (warmup 500)
- Effective batch size: 16 (per-device batch size 2, gradient accumulation 8)
- Mixed precision: fp16, with gradient checkpointing
- Generation max length: 225

## Evaluation results

Evaluated on the 1,547-utterance NICT-Tib1 test set under five metrics: Character Error Rate (CER), Syllable Error Rate (SER), and three Segmented Word Error Rate (SWER) variants using different automatic word segmenters (BoTok-SWER, BERT-SWER, and Gem-SWER, the last computed on a 500-utterance subset for API cost reasons). All values are micro-averaged. See the paper for full bootstrap confidence intervals.

| Metric | Value |
|---|---|
| CER (micro) | 0.7587 |
| SER (micro) | 0.9824 |
| BoTok-SWER (micro) | 1.0849 |
| BERT-SWER (micro, full 1,547-utt. test set) | 0.544 |
| Gem-SWER (micro, 500-utt. subset) | 1.1739 |



### Full model family comparison (standard fine-tuning)

| Model | Repo | CER | SER | BoTok-SWER | BERT-SWER | Gem-SWER |
|---|---|---|---|---|---|---|
| HuBERT Base | [tibetan-asr-nict-tib1-hubert-base](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-hubert-base) | 0.1352 | 0.3690 | 0.4477 | 0.161 | 0.9653 |
| Wav2Vec 2.0 Base | [tibetan-asr-nict-tib1-wav2vec2-base](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-wav2vec2-base) | 0.0745 | 0.2152 | 0.2747 | 0.097 | 0.7447 |
| Whisper Tiny | [tibetan-asr-nict-tib1-whisper-tiny](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-whisper-tiny) | 0.1560 | 0.2351 | 0.2975 | 0.118 | 0.6759 |
| Whisper Base | [tibetan-asr-nict-tib1-whisper-base](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-whisper-base) | 0.1417 | 0.2083 | 0.2600 | 0.105 | 0.6314 |
| **Whisper Small** | [tibetan-asr-nict-tib1-whisper-small](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-whisper-small) | **0.1185** | **0.1692** | **0.2042** | **0.086** | **0.5337** |


> [!WARNING]
> **This is a LoRA/QLoRA fine-tuned checkpoint. In the paper's benchmark, all LoRA and QLoRA configurations showed catastrophic word-level degradation** relative to full fine-tuning of the same base model, despite retaining partial character-level accuracy. This repo is published for reproducibility of that (negative) result, not as a recommended deployment artifact. If you need a usable Tibetan ASR model, use the [standard fine-tuned checkpoint](https://huggingface.co/billingsmoore/tibetan-asr-nict-tib1-whisper-tiny) instead.

## How to use

```python
from transformers import WhisperForConditionalGeneration, WhisperProcessor
from peft import PeftModel

base_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-tiny", device_map="auto")
model = PeftModel.from_pretrained(base_model, "billingsmoore/tibetan-asr-nict-tib1-whisper-tiny-lora")
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny", language="bo", task="transcribe")

# generate as usual, e.g.:
# inputs = processor(audio, sampling_rate=16000, return_tensors="pt")
# predicted_ids = model.generate(**inputs)
# transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)
```

## Limitations

- Trained and evaluated only on read-speech, modern **Lhasa (Central) Tibetan** news recordings; performance on Amdo or Kham dialects, Classical Tibetan, or spontaneous/conversational speech is unknown.
- The test set contains only 3 speakers (by design, to prevent speaker leakage), which limits how confidently results generalize across speaker, age, and prosodic variation.
- CER alone understates errors that matter at the word level for Tibetan; see the paper for why SER and SWER are necessary complements when judging output quality.
- LoRA/QLoRA adapters showed severe sequence-level collapse in this study (SWER often exceeding 1.0, i.e. worse than a random-length hypothesis) despite retaining moderate CER — treat this checkpoint as a research artifact documenting that failure mode, not as a usable transcription model.

## Citation

If you use this model, please cite the paper and the source NICT-Tib1 corpus:

```bibtex
@ARTICLE{11592371,
  author={Moore, Jacob and Li, Sheng and Lauren, Paula},
  journal={IEEE Access},
  title={Evaluating Tibetan ASR With Segmented Word Error Rate: Beyond Character-Level Metrics},
  year={2026},
  volume={14},
  number={},
  pages={101790-101805},
  keywords={Modeling;Automatic speech recognition;Error analysis;LoRa;Measurement;Ranking (statistics);Quantization (signal);Bit error rate;Standards;Training;Tibetan;automatic speech recognition;word error rate;low-resource language},
  doi={10.1109/ACCESS.2026.3709206}
}

@inproceedings{soky2022nict,
  title={Nict-tib1: A public speech corpus of lhasa dialect for benchmarking tibetan language speech recognition systems},
  author={Soky, Kak and Gong, Zhuo and Li, Sheng},
  booktitle={2022 25th Conference of the Oriental COCOSDA International Committee for the Co-ordination and Standardisation of Speech Databases and Assessment Techniques (O-COCOSDA)},
  pages={1--5},
  year={2022},
  organization={IEEE}
}
```

## License

Released under Apache License 2.0, inherited from the base model `openai/whisper-tiny`.