Hugging Face's logo

coml
/
maubert-feat

Safetensors
maubert
acoustic-unit-discovery
hubert
speech
Model card Files
xet
 Community
maubert-feat
File size: 2,810 Bytes
9f77069
14bda42
 
 
 
 
6d87aba
14bda42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9f77069
14bda42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6d87aba
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14bda42
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
---
library_name: maubert
license: mit
tags:
  - acoustic-unit-discovery
  - hubert
  - maubert
  - speech
language_bcp47:
  - ab
  - am
  - ba
  - be
  - bg
  - bn
  - ca
  - ckb
  - cs
  - cv
  - dv
  - el
  - eu
  - gn
  - ha
  - hi
  - hsb
  - hu
  - hy-AM
  - id
  - it
  - ja
  - ka
  - kk
  - kmr
  - ko
  - ky
  - lij
  - lt
  - ml
  - mn
  - mr
  - mt
  - myv
  - nan-tw
  - nl
  - pa-IN
  - pl
  - pt
  - ro
  - ru
  - rw
  - sk
  - sl
  - sq
  - sr
  - sv-SE
  - tk
  - tt
  - ug
  - ur
  - uz
  - vi
  - yo
  - yue
---

# MauBERT-feat

MauBERT-feat is a speech model adapting a pre-trained HuBERT base model to predict frame-level articulatory features on
a subset of 788 hours and 55 languages of Common Voice 16.1 leveraging the phone annotations from VoxCommunis. It was
introduced by the [MauBERT paper](https://aclanthology.org/2026.acl-long.24/) at ACL 2026.

You can load it with [`maubert`](https://github.com/bootphon/maubert):

```python
from maubert import MauBERTFeat

model = MauBERTFeat.from_pretrained("coml/maubert-feat")
```

Check out the [`maubert`](https://github.com/bootphon/maubert) repository if you are interested in further retraining
or fine-tuning, and the [corresponding paper](https://aclanthology.org/2026.acl-long.24/) for our results.

## Files:

- `model.safetensors` and `config.json`: HuggingFace Transformers checkpoint and config.

The manifests, phone alignments, language table and phone inventories used to train this model are released separately
in [`coml/vox-communis-artifacts`](https://huggingface.co/datasets/coml/vox-communis-artifacts). The inventories are
required to reduce the phone head when extracting features from the phone projection layer:

```python
from huggingface_hub import snapshot_download

data_dir = snapshot_download("coml/vox-communis-artifacts", repo_type="dataset")
# then pass f"{data_dir}/canonical_inventories.jsonl" as `inventory_file`
```

## Citing

```bibtex
@inproceedings{ortiztandazo-etal-2026-maubert,
    title = "{M}au{BERT}: Universal Phonetic Inductive Biases for Few-Shot Acoustic Units Discovery",
    author = "Ortiz Tandazo, Angelo  and
      Khentout, Manel  and
      Benchekroun, Youssef  and
      Hueber, Thomas  and
      Dupoux, Emmanuel",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.acl-long.24/",
    doi = "10.18653/v1/2026.acl-long.24",
    pages = "568--585",
    ISBN = "979-8-89176-390-6",
}
```