Loading datasets...
Release Date: 9/17/2026
Format: MP3
Size: 4.77 GB
A collection of read speech recordings in Abkhaz (Аԥсшәа).
Restrictions/Special Constraints
None provided.
Forbidden Usage
It is forbidden to attempt to determine the identity of speakers in the Common Voice datasets. It is forbidden to re-host or re-share this dataset.
Intended Use
This dataset is intended to be used for training and evaluating automatic speech recognition (ASR) models. It may also be used for applications relating to computer-aided language learning (CALL) and language or heritage revitalisation.
ab)This datasheet is for cv-corpus-27.0-2026-09-11 of the Mozilla Common Voice Scripted Speech dataset for Abkhaz [Аԥсшәа - ab]. The dataset contains 158625 clips representing 242.64 hours of recorded speech (207.84 hours validated) from 1303 speakers, recorded from a text corpus of 1,046,847 sentences.
| Code | Accent | Clips | Speakers |
|---|---|---|---|
| - | 75 (0.0%) | 4 (0.3%) |
The dataset includes the following self-declared age and gender distributions. A coverage summary is shown below each table.
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.
| Code | Gender | Clips | Speakers |
|---|---|---|---|
| male_masculine | Male, masculine | 11,052 (7.0%) | 81 (6.2%) |
| female_feminine | Female, feminine | 107,846 (68.0%) | 713 (54.7%) |
| transgender | Transgender | 13 (0.0%) | 1 (0.1%) |
| non-binary | Non-binary | - | - |
| do_not_wish_to_say | Prefer not to say | 1,042 (0.7%) | 6 (0.5%) |
| - | Unspecified | 38,672 (24.4%) | 577 (44.3%) |
Gender declared: 119,953 of 158,625 clips (75.6%), 726 of 1,303 speakers (55.7%)
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.
| Code | Age | Clips | Speakers |
|---|---|---|---|
| teens | Teens | 76,717 (48.4%) | 654 (50.2%) |
| twenties | Twenties | 22,667 (14.3%) | 161 (12.4%) |
| thirties | Thirties | 16,650 (10.5%) | 91 (7.0%) |
| fourties | Fourties | 8,148 (5.1%) | 52 (4.0%) |
| fifties | Fifties | 5,638 (3.6%) | 35 (2.7%) |
| sixties | Sixties | 3,146 (2.0%) | 22 (1.7%) |
| seventies | Seventies | 576 (0.4%) | 5 (0.4%) |
| eighties | Eighties | 394 (0.2%) | 2 (0.2%) |
| nineties | Nineties | 13 (0.0%) | 1 (0.1%) |
| - | Unspecified | 24,676 (15.6%) | 369 (28.3%) |
Age declared: 133,949 of 158,625 clips (84.4%), 934 of 1,303 speakers (71.7%)
Clip buckets
| Bucket | Clips |
|---|---|
| Validated | 135,878 (85.7%) |
| Invalidated | 22,483 (14.2%) |
| Other | 264 (0.2%) |
Training splits
| Split | Clips |
|---|---|
| Train | 97,639 (71.9%) |
| Dev | 14,111 (10.4%) |
| Test | 14,159 (10.4%) |
Training split coverage: 125,909 of 135,878 validated clips (92.7%)
The dataset contains 135878 validated, 22483 invalidated, and 264 unresolved clips. The average clip duration is 5.507 seconds.
Large-scale Abkhaz voice data collection campaign on Mozilla Common Voice
As of December 2025, the Abkhaz language dataset on the Mozilla Common Voice platform contained 104 hours of recorded speech, with a 65% validation rate. At that stage, 426 participants had contributed recordings. Starting in December 2025, by agreement with the Ministry of Education of the Republic of Abkhazia and with the participation of Abkhaz State University, the Shinkuba Abkhaz Language Development Foundation launched a large-scale campaign to expand the Abkhaz speech dataset on the Common Voice platform. During this period, specialists of the Foundation organized recording sessions across multiple regions of Abkhazia. Participants included students and professors of the university, high school students, school teachers, and volunteers from different age groups. As a result of this campaign, more than 150 additional hours of speech recordings were collected. By early March 2026, the Abkhaz dataset on Mozilla Common Voice reached a total of 254 hours of recorded speech, with 86% of the recordings validated. More than 800 new contributors participated in the campaign during this period.
Participation of educational institutions
The campaign involved numerous educational institutions throughout Abkhazia.
Sukhum
Ochamchira District
Gulrypsh District
Gudauta District
Gagra District
Tkvarchal
Abkhaz State University also played a major role in the project. Recording sessions initially started at the Faculty of Philology and later expanded to other faculties, involving approximately 200 students and professors.
Scale of public participation
Overall, more than 800 new contributors joined the project during this stage of data collection. Considering the size of the Abkhaz population, the campaign reached a remarkable level of public participation: more than 1% of the Abkhaz population contributed their voices to the dataset. Such a level of public engagement in building a speech dataset is extremely rare and represents a remarkable example of community participation in the digital development of a minority language.
Documentation and public outreach
All recording sessions were documented through photo and video coverage. The project team recorded each school visit and community event where speech recordings were collected. Project updates and documentation can be found on the Foundation’s public channels:
Instagram https://www.instagram.com/shinkuba_apsua_language
Facebook https://www.facebook.com/share/1HGjWEpK9m/?mibextid=wwXIfr
Television interview and report https://youtu.be/qUQKo-cMtJw?si=O9bYwB8XETSCBnmC
Significance of the campaign
The expanded Abkhaz speech dataset represents an important milestone in the digital presence of the Abkhaz language and significantly strengthens the available resources for future speech technologies and AI-based language tools.
Validated sentences: 1,046,833
| Category | Count |
|---|---|
| Unvalidated sentences | 14 |
| Pending sentences | 9 |
| Rejected sentences | 5 |
| Reported sentences | 346 |
The corpus contains 1,046,847 sentences: 1,046,833 validated and 14 unvalidated (9 pending review, 5 rejected), with 346 reported for review.
There follows a randomly selected sample of five sentences from the corpus.
| Source | Sentences |
|---|---|
| batch3 | 664,250 (63.5%) |
| batch2 | 286,152 (27.3%) |
| batch1 | 91,358 (8.7%) |
| Other | 5,073 (0.5%) |
| Code | Domain | Clips | Speakers |
|---|---|---|---|
| general | General | - | - |
| agriculture_food | Agriculture and Food | 1 (0.0%) | 1 (0.1%) |
| automotive_transport | Automotive and Transport | 2 (0.0%) | 2 (0.2%) |
| finance | Finance | - | - |
| service_retail | Service and Retail | - | - |
| healthcare | Healthcare | - | - |
| history_law_government | History, Law and Government | 2 (0.0%) | 2 (0.2%) |
| media_entertainment | Media and Entertainment | - | - |
| nature_environment | Nature and Environment | 1 (0.0%) | 1 (0.1%) |
| news_current_affairs | News and Current Affairs | - | - |
| technology_robotics | Technology and Robotics | - | - |
| language_fundamentals | Language Fundamentals | - | - |
Each row of a tsv file represents a single audio clip, and contains the following information:
client_id - hashed UUID of a given userpath - relative path of the audio filesentence - the sentence to be read aloudsentence_id - unique identifier for the sentencesentence_domain - domain classification(s) of the sentenceup_votes - number of people who said audio matches the textdown_votes - number of people who said audio does not match textage - age of the speaker1gender - gender of the speaker1accents - accents of the speaker1variant - variant of the language1locale - locale code of the languagesegment - if sentence belongs to a custom dataset segment, it will be listed herevalidated_sentences.tsvThe validated_sentences.tsv file contains one row per validated sentence in the text corpus:
sentence_id - unique identifier for the sentencesentence - the sentence textvariant - the variant of the languagesentence_domain - the domain(s) the sentence belongs tosource - the source the sentence was collected fromis_used - whether the sentence is still in circulation for recordingclips_count - number of clips recorded for this sentenceunvalidated_sentences.tsvThe unvalidated_sentences.tsv file contains one row per unvalidated sentence in the text corpus:
sentence_id - unique identifier for the sentencesentence - the sentence textvariant - the variant of the languagesentence_domain - the domain(s) the sentence belongs tosource - the source the sentence was collected fromup_votes - number of upvotes the sentence receiveddown_votes - number of downvotes the sentence receivedstatus - current status of the sentence (pending or rejected)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.
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