Automatic Speech Recognition
Welsh
kaldi
vosk
welsh
cymraeg
speech-recognition
File size: 8,232 Bytes
c1adedb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
125
126
127
kaldi-cy 2610, released October 2026
Copyright 2026 Bangor University

Developed by the Language Technologies Unit (Uned Technolegau Iaith), Bangor University.

Licensed under the Apache License, Version 2.0 (the "License"). You may obtain a copy of the
License at http://www.apache.org/licenses/LICENSE-2.0

NAMES
This licence grants no permission to use the names "Bangor University", "Prifysgol Bangor",
"Language Technologies Unit", "Uned Technolegau Iaith", "techiaith", the University's logos, or
the names of its products or staff, except to describe the origin of the model as this NOTICE
does.

WHAT THIS MODEL IS
This is a research model, trained on the data and evaluated by the methods described in its
model card. It is not a certified product and is not a substitute for a qualified translator,
transcriber, clinician or other professional.

AUTOMATIC PROCESSING MAKES MISTAKES
Output from this model, whether a translation, a transcript, generated text or speech, or any
other result, can be wrong while appearing fluent and confident. Known failure modes include
omitted or inserted words, incorrect names, numbers, dates and negations, and reduced accuracy
on dialects, noisy audio or subjects outside the training data.

Prifysgol Bangor University therefore advises that output is treated as provisional and is
checked and corrected by a competent person before it is relied on as a record, used in any
decision affecting a person, or shared or published. Where the model is used interactively or in
real time (for example live captions, dictation, suggestions in an editor or read-aloud), users
should be told that what they see or hear is automatic and may be wrong. The University asks
anyone who builds on or redistributes this model to pass this warning on to their own users.

NOT FIT FOR
The model has not been evaluated for, and should not be used in, medical, legal, financial,
safety-critical or emergency settings, or in any decision affecting a person's rights, health,
liberty or livelihood, unless a qualified person reviews every output.

USE IN WALES: EQUAL TREATMENT OF WELSH AND ENGLISH
This model was developed to help organisations provide services in Welsh that are as good as
those they provide in English, not to replace Welsh-language provision. Under section 1 of the
Welsh Language (Wales) Measure 2011 (2011 nawm 1), Welsh has official status in Wales, and that
status is given legal effect by, among other things, the treatment of Welsh no less favourably
than English (s. 1(2)(b)) and by standards of conduct relating to the use of Welsh (s. 1(3)(e)).
These standards are made and enforced under Part 4 of the Measure. Bodies still operating a
Welsh language scheme under Part II of the Welsh Language Act 1993 (c. 38) must give effect to
the principle in section 5 of that Act that, in the conduct of public business and the
administration of justice in Wales, Welsh and English should be treated on a basis of equality.
Prifysgol Bangor University asks anyone deploying this model in Wales to consider whether the
intended use meets those requirements. Offering automated Welsh where English is served by
people, publishing machine translation into Welsh without the review given to English text, or
otherwise using this technology to replace rather than support Welsh-language provision is
contrary to the purpose for which the model was released. Which duties apply depends on the body
and the regulations or scheme that cover it, and responsibility for compliance rests with the
body deploying the model. See the Welsh Language Commissioner's regulatory policy statement on
artificial intelligence and the Welsh language, Technology, Websites and Software: Welsh
Language Considerations, and the advice document Bilingual drafting, translation and using Welsh
face to face, all linked from the model card. The legislation is at
legislation.gov.uk/mwa/2011/1 and legislation.gov.uk/ukpga/1993/38.

FURTHER INFORMATION AND ERROR REPORTS
Training data, evaluation results and known limitations: https://huggingface.co/techiaith/kaldi-cy
Report errors or concerns to: techiaith@bangor.ac.uk

PRONUNCIATION DICTIONARY
The decoding graph (HCLG.fst) and word list (words.txt) in this bundle were built with
pronunciations from Geiriadur Ynganu Bangor, the Bangor Pronouncing Dictionary,
https://github.com/techiaith/geiriadur-ynganu-bangor, redistributed here under its own
licence, which requires the following to be kept with any binary redistribution:

  Copyright (C) 2025 Prifysgol Bangor University. All rights reserved.

  Redistribution and use in source and binary forms, with or without modification, are
  permitted provided that the following conditions are met:

  1. Redistributions of source code must retain the above copyright notice, this list of
     conditions and the following disclaimer. The contents of this file are deemed to be
     source code.

  2. Redistributions in binary form must reproduce the above copyright notice, this list
     of conditions and the following disclaimer in the documentation and/or other
     materials provided with the distribution.

  THIS SOFTWARE IS PROVIDED BY PRIFYSGOL BANGOR UNIVERSITY ``AS IS'' AND ANY EXPRESSED OR
  IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
  PRIFYSGOL BANGOR UNIVERSITY NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT,
  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

BUILT WITH
Trained with Kaldi (Apache License 2.0) and pocolm (Apache License 2.0), and packaged
for Vosk (Apache License 2.0), following the vosk-cymraeg recipe by Gweltaz
Duval-Guennoc, Preben Vangberg and Sasha Wanasky (MIT License, Cymru-Breizh Agile Cymru
Project, https://github.com/Cymru-Breizh-Agile-Cymru-Project/vosk-cymraeg). None of
their code is included in this bundle.

TRAINING DATA ATTRIBUTIONS
The Apache License applies to the model weights, code and documentation. It does not license the
data the model was trained on, which is not distributed with the model and remains under the
terms of its original providers, acknowledged here as those terms require.

Dedicated to the public domain under CC0 1.0 Universal
(https://creativecommons.org/publicdomain/zero/1.0/). CC0 asks for no attribution; the sources
are listed so that the training data can be traced to its datasheet.
Banc Trawsgrifiadau Bangor (CC0 1.0), collected by Language Technologies Unit, Bangor
  University. Collection process, consent terms and contact:
  https://huggingface.co/datasets/techiaith/banc-trawsgrifiadau-bangor
Corpws CLlLC / WLGA (CC0 1.0), collected by Language Technologies Unit, Bangor University.
  Collection process, consent terms and contact:
  https://huggingface.co/datasets/techiaith/corpws-clllc-wlga
Mozilla Common Voice 23.0, Welsh (CC0 1.0), collected by Mozilla Foundation and the volunteers
  who contributed their voices. Collection process, consent terms and contact:
  https://commonvoice.mozilla.org/cy/datasets. Obtained from Mozilla under their terms for
  Common Voice data and held in a private working copy; it is not redistributed with or by this
  model.
Lleisiau ARFOR (CC0 1.0), collected by Cymen, with the Language Technologies Unit, Bangor
  University. Collection process, consent terms and contact:
  https://huggingface.co/datasets/cymen-arfor/lleisiau-arfor
Mozilla Common Voice, Welsh, re-segmented with voice activity detection (CC0 1.0), collected by
  Mozilla Foundation and the volunteers who contributed their voices. Collection process,
  consent terms and contact: https://commonvoice.mozilla.org/cy/datasets. Derived from Common
  Voice data obtained from Mozilla under their terms and held in a private working copy; it is
  not redistributed with or by this model.