---
license: other
license_name: pantagruel-research-license
license_link: https://huggingface.co/ina-foss/ssl-audio-1k-base/blob/main/LICENSE.md
extra_gated_prompt: >-
  You agree to use the model according to its
  [license](https://huggingface.co/ina-foss/ssl-audio-1k-base/blob/main/LICENSE.md).
extra_gated_fields:
  Research institute: text
  Institutional email address: text
  Country: country
  I agree to use this model for non-commercial research use only: checkbox
  I have read, and I agree to use the model according to its license: checkbox
language:
- fr
pipeline_tag: feature-extraction
library_name: transformers
tags:
- data2vec2
- JEPA
- speech
- fairseq
---

# SSL Audio 1k-base

This model is pretrained on 1,000 hours of audio content from INA, sampled following the `base` setting, as described in our LREC 2026 paper "Data Selection Effects on Self-Supervised Learning of Audio Representations for French Audiovisual Broadcasts".

| **Link to model** | **Pretraining data selection** |
|-------------------|--------------------|
| [ssl-audio-1k-base](https://huggingface.co/ina-foss/ssl-audio-1k-base) | Random sample of 1,000h |
| [ssl-audio-1k-no_music](https://huggingface.co/ina-foss/ssl-audio-1k-no_music) | Samples not containing musics |
| [ssl-audio-1k-only_speech](https://huggingface.co/ina-foss/ssl-audio-1k-only_speech) | Samples only composed of speech |
| [ssl-audio-1k-only_fr](https://huggingface.co/ina-foss/ssl-audio-1k-only_fr) | Samples only composed of French content |
| [ssl-audio-1k-gender](https://huggingface.co/ina-foss/ssl-audio-1k-gender) | Samples with a balanced proportion of male and female speech |
| ssl-audio-1k-duplicates | Samples with duplicates content. This model is not released. |

The features generated by these models have been used for [Voice Activity Detection (VAD)](https://huggingface.co/collections/ina-foss/ssl-vad) and [music detection](https://huggingface.co/collections/ina-foss/ssl-music-detection)
For detailed information about training and results associated with this model, please refer to our publication.
Along with the [Tensorboard training metrics](https://huggingface.co/ina-foss/ssl-audio-1k-base/tensorboard), we release the hyperparameters.

## Usage

```python
import librosa
from transformers import AutoModel, AutoFeatureExtractor

# loading the audio file, need to be sampled at 16kHz
audio, sr = librosa.load('/path/to/your/audio/file.wav', sr=16000)

# loading the feature extractor and SSL model
model_name = 'ina-foss/ssl-audio-1k-base'
model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
processor = AutoFeatureExtractor.from_pretrained(model_name)
model.eval()

inputs = processor(audio, sampling_rate=sr, return_tensors="pt")

# extract features
with torch.no_grad():
    outputs = model(**inputs)
```

## License and citation

The model is distributed using the [pantagruel-research-license](https://huggingface.co/ina-foss/ssl-audio-1k-base/blob/main/LICENSE.md).

If you use this model or find it useful in your research, publications, or applications, please cite the following work:
```
@inproceedings{pelloin2026lrec,
  author =       "Pelloin, Valentin and Bekkali, Lina and Dehak, Reda and Doukhan, David",
  year =         "2026",
  title =        "Data Selection Effects on Self-Supervised Learning of Audio Representations for French Audiovisual Broadcasts",
  booktitle={Fifteenth International Conference on Language Resources and Evaluation (LREC 2026)},
  address = "Palma, Mallorca, Spain",
  publisher = "European Language Resources Association",
}
```