---
license: mit
language:
- en
library_name: transformers
pipeline_tag: automatic-speech-recognition
tags:
- whisper
- automatic-speech-recognition
- gumbel-beard
- children-speech
- low-resource
base_model: openai/whisper-small.en
---

# Whisper small.en — Gumbel-BEARD

This is an [openai/whisper-small.en](https://huggingface.co/openai/whisper-small.en)
model whose encoder was domain-adapted with **Gumbel-BEARD** — a self-supervised
continued-pretraining recipe — and then fine-tuned for automatic speech
recognition (ASR) on the [MyST](https://catalog.ldc.upenn.edu/LDC2021S05)
children's speech corpus.

On the MyST test set it achieves **8.5% WER**.


- **Base model:** `openai/whisper-small.en`
- **Task:** English automatic speech recognition
- **Adaptation:** self-supervised (transcript-free) encoder adaptation via
  Gumbel-BEARD, followed by supervised ASR fine-tuning
- **Code:** <your-github-repo-url>

## Usage

```python
from transformers import pipeline

asr = pipeline(
    "automatic-speech-recognition",
    model="Zilai2999/whisper-small.en-gumbel-beard",
)
print(asr("audio.wav")["text"])
```

Or with the model/processor directly:

```python
import torch, librosa
from transformers import WhisperForConditionalGeneration, WhisperProcessor

model_id = "Zilai2999/whisper-small.en-gumbel-beard"
processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)

audio, _ = librosa.load("audio.wav", sr=16000)
inputs = processor(audio, sampling_rate=16000, return_tensors="pt")
ids = model.generate(inputs.input_features)
print(processor.batch_decode(ids, skip_special_tokens=True)[0])
```

Audio is expected at 16 kHz.

## Training

- Self-supervised Gumbel-BEARD adaptation of the Whisper encoder (BEST-RQ +
  self-distillation with automatic Gumbel-softmax layer selection).
- Supervised ASR fine-tuning on the **MyST** children's speech corpus.

See the code repository for the full recipe and hyper-parameters.

## Results

| Test set | WER |
| --- | --- |
| MyST | **8.5%** |


## Citation

```bibtex
@inproceedings{gumbelbeard2026,
  title     = {Gumbel-BEARD: Automatic Layer Selection for Self-Supervised
               Adaptation of Whisper in Low-Resource Domains},
  booktitle = {Proc. Interspeech},
  year      = {2026},
}
```