Common Voice Scripted Speech 25.0 - Welsh

Cymraeg — Welsh (cy)

This datasheet is for cv-corpus-25.0-2026-03-09 of the Mozilla Common Voice Scripted Speech dataset for Welsh [Cymraeg - cy]. The dataset contains 116102 clips representing 158.55 hours of recorded speech (124.11 hours validated) from 1933 speakers, recorded from a text corpus of 118,059 sentences.

Language

The Welsh language (Cymraeg - /kəmˈraːiɡ/) belongs to the Celtic language family, specifically the Brythonic branch of Insular Celtic. Welsh evolved in the 6th century, alongside Breton and Cornish, from Common Brittonic, the common language previously spoken throughout Great Britain during the Iron Age and Roman period. The language has experienced contact with English since medieval times, leading to extensive lexical borrowing and gradual language shift intensified by socioeconomic factors and educational policies. It's resilience however stems from concentrated communities, intergenerational transmission, cultural traditions like the Eisteddfod, and religious revival movements. Strategic activism by speakers from the 1960s onwards has transformed Welsh from a declining minority language into one with official legal status. The most recent census shows 538,300 speakers (17.8% of Wales' population), though other surveys suggest higher figures of up to 862,700 speakers. Legislation by the Senedd (Welsh Parliament) supports revitalization through education and technology, with the goal of achieving one million speakers and doubled daily usage by 2050.

Welsh language speech varies significantly across proficiency levels (from native speakers and learners to passive understanders), regional dialects (mainly northern and southern, plus Patagonian Welsh), and registers (formal literary Welsh versus more informal colloquial spoken forms that sometimes incorporates code-switching). Speaker competence ranges widely, with some demonstrating full fluency across all domains while others show more limited or context-specific abilities.

Variants

Accents

Demographic information

The dataset includes the following self-declared age and gender distributions. A coverage summary is shown below each table.

Gender

Self-declared gender information. The table shows clip and speaker counts with percentages. Speakers who did not declare a gender are listed as Unspecified. A dash (-) indicates zero.

Gender declared: 66,426 of 116,102 clips (57.2%), 429 of 1,933 speakers (22.2%)

Age

Self-declared age information. The table shows clip and speaker counts with percentages. Speakers who did not declare an age are listed as Unspecified. A dash (-) indicates zero.

Age declared: 66,720 of 116,102 clips (57.5%), 442 of 1,933 speakers (22.9%)

Data splits for modelling

Clip buckets

Training splits

Training split coverage: 18,858 of 90,884 validated clips (20.7%)

The dataset contains 90884 validated, 4472 invalidated, and 20746 unresolved clips. The average clip duration is 4.916 seconds.

Text corpus

Validated sentences: 118,044

The corpus contains 118,059 sentences: 118,044 validated and 15 unvalidated (1 pending review, 14 rejected), with 176 reported for review.

Writing system

Welsh employs a Latin alphabet of 29 letters, including eight digraphs (ch, dd, ff, ng, ll, ph, rh, th) treated as single letters for alphabetization, and treats "w" and "y" as vowels. The orthography is largely phonetic with predictable sound-letter correspondences. Loanwords, especially from English, are occasionally adapted to Welsh spelling conventions.

Symbol table

The 29 letters used for alphabetization in Welsh are:

a b c ch d dd e f ff g ng h i j l ll m n o p ph r rh s t th u w y

The letter j is borrowed from English to represent the borrowed /dʒ/ consonant phoneme.

Welsh also uses diacritical marks on vowels (considered variants of their base letters, not separate letters for alphabetization):

With circumflex ('to bach'): â ê î ô û ŵ ŷ

With acute accent: á é í ó ú ẃ ý

With grave accent: à è ì ò ù ẁ ỳ

With diaeresis: ä ë ï ö ü ẅ ÿ

The traditional system lacks letters k, q, v, x, z, although these can appear in proper nouns and technical terms.

Sample

There follows a randomly selected sample of five sentences from the corpus.

1. Mae cyfanswm sgoriau uwch yn dangos symptomau iselder mwy difrifol.

2. Caiff y pentref ei lywodraethu gan ffurf Maer-Cyngor ar lywodraeth.

3. Mae f’athrawes Gatalaneg i yn mynd ar gwrs Cymraeg yr haf ’ma.

4. Dywed rhai cipwyr fod hela copaon yn gymhelliad i barhau i gyrraedd copaon newydd.

5. Yn yr amser hwnnw, mae wyth o reolwyr gwahanol wedi cael eu penodi.

Sources

Text domains

Fields

Clips

Each row of a tsv file represents a single audio clip, and contains the following information:

• client_id - hashed UUID of a given user

• path - relative path of the audio file

• text - supposed transcription of the audio

• up_votes - number of people who said audio matches the text

• down_votes - number of people who said audio does not match text

• age - age of the speaker1

• gender - gender of the speaker1

• accents - accents of the speaker1

• variant - variant of the language1

• segment - if sentence belongs to a custom dataset segment, it will be listed here

• prompt_upvotes - number of upvotes the sentence prompt received

• prompt_reports - number of reports the sentence prompt received

• is_edited - whether the clip's transcription has been edited

validated_sentences.tsv

The validated_sentences.tsv file contains one row per validated sentence in the text corpus:

• sentence_id - unique identifier for the sentence

• sentence - the sentence text

• variant - the variant of the language

• sentence_domain - the domain(s) the sentence belongs to

• source - the source the sentence was collected from

• is_used - whether the sentence is still in circulation for recording

• clips_count - number of clips recorded for this sentence

unvalidated_sentences.tsv

The unvalidated_sentences.tsv file contains one row per unvalidated sentence in the text corpus:

• sentence_id - unique identifier for the sentence

• sentence - the sentence text

• variant - the variant of the language

• sentence_domain - the domain(s) the sentence belongs to

• source - the source the sentence was collected from

• up_votes - number of upvotes the sentence received

• down_votes - number of downvotes the sentence received

• status - current status of the sentence (pending or rejected)

Get involved

Community links

• Common Voice translators on Pontoon

• Common Voice Communities

Discussions

• Common Voice on Matrix

• Common Voice on Discourse

• Common Voice on Discord

• Common Voice on Telegram

Contribute

• Speak

• Write

• Listen

• Review

Licence

This dataset is released under the Creative Commons Zero (CC-0) licence. By downloading this data you agree to not determine the identity of speakers in the dataset.

Footnotes

1. For a full list of age, gender, and accent options, see the demographics spec. These will only be reported if the speaker opted in to provide that information. ↩ ↩2 ↩3 ↩4