Views
No views yet
1# Clone the repository
2git clone [https://huggingface.co/mustafa-ozan-duman/wavlm-transformer-mos-english](https://huggingface.co/mustafa-ozan-duman/wavlm-transformer-mos-english)
3cd wavlm-transformer-mos-english
4
5# Install requirements
6pip install -r requirements.txt
7
8🚀 How to Generate MOS Labels
9We have provided a dedicated script, predict_folder.py, to handle batch generation. This script will scan a directory for .wav files, convert them to the required 16kHz format automatically, and save the predictions to a CSV file.
10
11
12Basic Usage:
13Run the following command in your terminal:
14
15Bash
16python predict_folder.py --dir path/to/your/audio_folder --out my_predictions.csv
17
18Script Arguments:
19--dir: (Required) Path to the directory containing your audio files.
20
21--out: (Optional) Name of the output CSV file (Default: mos_predictions.csv).
22
23
24
25
26📂 Repository Structure
27best_model.pth: The trained model weights (1.33 GB).
28
29model.py: The WavLM + Transformer architecture definition.
30
31dataset.py: Audio preprocessing and loading logic.
32
33predict_folder.py: The main script for generating MOS labels.
34
35requirements.txt: Exact library versions used during development.
36
37📝 Technical Details
38Sampling Rate: The model operates at 16,000 Hz. Audios with different sampling rates are automatically resampled by the prediction script.
39
40Input Format: Mono PCM Wav.
41
42Output: A CSV file with two columns: wav_filename and predicted_mos.
43
44Author: Mustafa Ozan Duman
45
46Affiliation: Research Assistant, Bursa Uludag University
47
48Contact: mustafaduman@uludag.edu.tr
49
50🎓 Citation
51If you use this model or code in your research, please cite our paper:
52
53Text Citation:
54
55Mustafa Ozan Duman and Ahmet Emir Dirik. "Evaluating SSL and ViViT Architectures for Cross-Corpus Audio MOS Prediction via LODO Validation." arXiv preprint arXiv:2607.10146 (2026).
56
57BibTeX:
58
59Kod snippet'i
60@misc{duman2026evaluatingsslvivitarchitectures,
61 title={Evaluating SSL and ViViT Architectures for Cross-Corpus Audio MOS Prediction via LODO Validation},
62 author={Mustafa Ozan Duman and Ahmet Emir Dirik},
63 year={2026},
64 eprint={2607.10146},
65 archivePrefix={arXiv},
66 primaryClass={eess.AS},
67 url={[https://arxiv.org/abs/2607.10146](https://arxiv.org/abs/2607.10146)},
68}