ProxyMOS

ProxyMOS is a lightweight Mean Opinion Score (MOS) prediction model built on top of the OmniASR W2V 300M encoder (Wav2Vec2 architecture via fairseq2).

Given any audio file, the model returns a single scalar speech quality score on the MOS scale. It is the distilled model from the paper "ProxyMOS: Training Speech Quality Models via Ensemble-Derived MOS Targets" (Borodin & Trokunov, MTUCI, 2026).

---

Architecture

OmniASR-W2V-300M (encoder, fine-tuned with reduced lr)
        ↓
AttentiveStatsPooling  ->  [mean || std]  (dim: 2 x 1024 = 2048)
        ↓
Linear(2048 -> 1024) -> GELU -> Linear(1024 -> 1)
        ↓
     MOS score (scalar)

• Encoder: omniASR_W2V_300M — a 300M-parameter Wav2Vec2-style encoder instantiated via fairseq2.
• Weights: best_model_full.pt contains the trained ProxyMOS state dict, including the encoder and MOS head weights. No separate omniASR-W2V-300M.pt checkpoint is required.
• Pooling: Attentive Statistics Pooling — computes attention-weighted mean and standard deviation over the time dimension.
• Head: Two-layer MLP with GELU activation and a scalar output.

---

Repository files

File	Description
config.json	Model metadata used by Hugging Face Hub tooling
best_model_full.pt	Trained ProxyMOS state dict, including encoder and MOS head weights (~1.28 GB)
inference_model.py	Model architecture + inference script

Note: inference uses fairseq2 to instantiate the omniASR_W2V_300M architecture, then loads all trained weights from best_model_full.pt. You do not need a separate omniASR-W2V-300M.pt file.

---

Quick start

1. Clone the repository

git lfs install
git clone https://huggingface.co/lab260/ProxyMos
cd ProxyMos

2. Install dependencies

Requires Python 3.10+.

pip install torch torchaudio
pip install fairseq2
pip install accelerate scipy scikit-learn tqdm

GPU: if CUDA is available, the model will use it automatically.

3. Run inference

python inference_model.py /path/to/your/audio.wav

Example:

python inference_model.py /home/ae_samples_ae_finetuned_vocoder_segment_28.wav

Output:

MOS: 3.8721

---

Python API

import torch
from inference_model import load_model, predict_mos

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = load_model("best_model_full.pt", device)

score = predict_mos(model, "/path/to/audio.wav", device)
print(f"MOS: {score:.4f}")

Batch processing

from pathlib import Path

audio_dir = Path("/path/to/audio_folder")

for wav_file in sorted(audio_dir.glob("*.wav")):
    score = predict_mos(model, str(wav_file), device)
    print(f"{wav_file.name}: {score:.4f}")

---

Model Performance

URGENT Dataset

Model	Pearson	Spearman	Kendall	RMSE	MAE
Proxymos	0.802	0.799	0.617	0.474	0.370
Proxymosonnx	0.779	0.778	0.591	0.502	0.393
Proxymosonnx16	0.779	0.778	0.591	0.502	0.393
Whisqa	0.780	0.773	0.588	0.502	0.397
Distillmos	0.759	0.748	0.568	0.525	0.411
UTMOS	0.721	0.748	0.560	0.565	0.450
XLS	0.743	0.745	0.561	0.542	0.426
MosNet	0.279	0.110	0.073	0.908	0.702
HuBERT	-0.005	-0.002	-0.001	1.073	0.855
NISQA	0.028	-0.027	-0.021	1.055	0.863
Dnsmos	-0.010	-0.041	-0.028	1.075	0.876

MOS260 Dataset

Model	Pearson	Spearman	Kendall	RMSE	MAE
Proxymosonnx	0.700	0.647	0.481	0.883	0.687
Proxymosonnx16	0.700	0.647	0.481	0.883	0.687
Proxymos	0.691	0.636	0.474	0.897	0.692
Distillmos	0.670	0.613	0.456	0.927	0.712
UTMOS	0.622	0.555	0.403	0.991	0.781
XLS	0.583	0.473	0.337	1.041	0.834
Whisqa	0.576	0.466	0.332	1.050	0.841
NISQA	0.261	0.226	0.158	1.386	1.105
Dnsmos	0.075	0.056	0.039	1.551	1.254
HuBERT	-0.003	-0.009	-0.006	1.614	1.294
MosNet	-0.210	-0.209	-0.146	1.773	1.384

Requirements

Package	Version
Python	≥ 3.10
PyTorch	≥ 2.0
torchaudio	compatible with torch
fairseq2	≥ 0.3
accelerate	≥ 0.20
scipy	≥ 1.10
scikit-learn	≥ 1.2

---

Technical details

• Audio is automatically resampled to 16,000 Hz.
• Stereo input is averaged to mono before processing.
• Input is normalized to shape [B, T] internally.
• Inference runs correctly on CPU when no GPU is available.

---

Results

Benchmark	Spearman ρ	Pearson r	RMSE
URGENT	0.802	0.806	0.471
mos260 (Russian TTS)	0.647	0.700	0.883

ProxyMOS outperforms all individual teacher models (WhiSQA, DistillMOS, UTMOS, XLS-R) on both benchmarks.

---

Citation

@article{borodin2026proxymos,
  title     = {ProxyMOS: Training Speech Quality Models via Ensemble-Derived MOS Targets},
  author    = {Borodin, Kirill and Trokunov, Maksim},
  year      = {2026},
  institution = {MTUCI, Moscow}
}

---

License

Please refer to the fairseq2 and omnilingual-asr repositories for encoder licensing terms.

Contact

• Email: kborodin.research@gmail.com
• Telegram: @korallll_ai