Common Voice Scripted Speech 25.0 - Kabardian

Адыгэбзэ (Къэбэрдей) — Kabardian (kbd)

This datasheet is for cv-corpus-25.0-2026-03-09 of the Mozilla Common Voice Scripted Speech dataset for Kabardian [Адыгэбзэ (Къэбэрдей) - kbd]. The dataset contains 168564 clips representing 291.4 hours of recorded speech (272.16 hours validated) from 347 speakers, recorded from a text corpus of 32,644 sentences.

Language

Kabardian (/kəˈbɑːrdiən/, also known as East Circassian) is a Northwest Caucasian language spoken by the eastern subgroups of Adygheans (Circassians). Identified by the ISO 639-3 code kbd, it belongs to the Northwest Caucasian language family and is widely considered to be the eastern dialect of Adyghe. Kabardian is also one of the official languages of the Kabardino-Balkarian Republic and the Karachay-Cherkess Republic, both federal subjects of the Russian Federation.

The Kabardian (East Circassians) population is estimated at 500,000 to 750,000, yet only around 300,000 are believed to actively speak the language. Kabardians are primarily located in the Kabardino-Balkaria and Karachay-Cherkessia republics of Russia, with additional diaspora populations in Turkey, Jordan, Syria, and other countries due to historical migrations. The language includes several dialects, with the Baksan and Malka varieties being the most prominent. However, only a limited portion of the Kabardian population is considered literate in the language.

There is no definitive and official publication on this matter. Data published by organizations such as Wikipedia, Etnologue, and the Joshua Project do not reflect reality. Many Kabardians living in both the Caucasus and the Diaspora identify themselves as Adyghe unless specifically asked. Therefore, to obtain reliable results, researchers must be well-versed in the Adyghe tribes and their culture.

Variants

The following table presents the general language variants of the Kabardian language used in the Common Voice project.

Sentences in the MCV project are provided in Cyrillic and Turkish transliteration to support non-literate users from the Turkish diaspora. With "literacy" we mean being able to read the Kabardian alphabet.

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: 72,240 of 168,564 clips (42.9%), 133 of 347 speakers (38.3%)

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: 145,151 of 168,564 clips (86.1%), 218 of 347 speakers (62.8%)

Data splits for modelling

Clip buckets

Training splits

Training split coverage: 28,996 of 157,434 validated clips (18.4%)

The dataset contains 157434 validated, 2763 invalidated, and 8367 unresolved clips. The average clip duration is 6.223 seconds.

Text corpus

The text corpus for the Kabardian language in the Common Voice project consists of a diverse collection of sentences contributed by native speakers. These sentences cover a wide range of topics, including everyday conversations, cultural references, and various domains such as literature, news articles, conversational texts, technology, health, and education. The maximum length of Kabardian sentences in this dataset is limited to 15 words and 140 characters due to restrictions in the Common Voice interface. However, Kabardian sentences can naturally be much longer. In this dataset, the average length has been aimed to be maintained at around 6–8 words and 80–100 characters to reflect the typical structure and complexity of the Kabardian language. The corpus has been curated to ensure a broad representation of vocabulary and grammatical constructs, making it suitable for training robust speech recognition models.

Validated sentences: 30,253

The corpus contains 32,644 sentences: 30,253 validated and 2,391 unvalidated (964 pending review, 1,427 rejected), with 3,995 reported for review.

Writing system

The Kabardian language uses the Cyrillic script with some additional letters to represent specific sounds in the language. The alphabet consists of 59 letters, including 33 standard Cyrillic letters and 26 additional letters unique to Kabardian. The writing system is phonetic, meaning that words are generally spelled as they are pronounced. Kabardian is written from left to right.

During the period when there was no Kabardian keyboard option, the recorded text content was written with the Russian keyboard as the closest alternative. The characters in the Kabardian alphabet defined by the Unicode Consortium as "Cyrillic Letter Palocka ('Ӏ': u04C0, utf-8)" and "Cyrillic Letter Small Palocka ('ӏ': u04CF, utf-8)" are not defined in this keyboard model. For this reason, since the early days of computers, when creating digital data in Adyghe or Kabardian, one of the characters "Latin Capital Letter i ('I': u0049, utf-8)", "Latin Small Letter L ('l': u006C, utf-8)" or "Digit One ('1': u0031, utf-8)", albeit rarely "Vertical Line ('|': u007C, utf-8)" was used randomly due to their visual similarities (Nemlioğlu, 2018). This situation emerges as an important factor that negatively affects data quality (Nemlioğlu,2025).

Symbol table

Main Alphabet: (Ordered by ascending) а э б в г {гу} {гъ} {гъу} д {дж} {дз} е ё ж {жь} з и й к {ку} {кӏ} {кӏу} {къ} {къу} {кхъ} {кхъу} л {лъ} {лӏ} м н о п {пӏ} р с т {тӏ} у ф {фӏ} х {ху} {хь} {хъ} {хъу} ц {цӏ} ч ш щ {щӏ} ъ ы ь ю я ӏ {ӏу}

Orthographic–Phonetic Table (IPA):

Auxiliary exemplar: The following letters are also present in the dataset, but are not part of the official Kabardian alphabet. They are used to represent three specific sounds in the different sub dialects.

{гь}: Voiced dorso-palatal plosive [ʄ] (e.g. гьанэ (ɟa:ne) [Some dialects of Kabardian] = джанэ [Kabardian]: shirt/dress) {кь}: Voiceless dorso-palato-velar plosive [kʲ] (e.g. кьыржын (kʲərʒən) [Some dialects of Kabardian] = чыржын [Kabardian]: a type of cookie made from corn flour ) {кӏь}: Glottalized voiceless dorso-palato-velar plosive [kʲʼ] (e.g. гьэдыкIьэ (ɟedəkʲ’e) [Some dialects of Kabardian] = джэдыкӏэ [Kabardian]: egg)

Prior to dataset utilization, the following transformations should be applied to resolve transcription inconsistencies related to dialectal phoneme representation:

г' -> гь чӏ -> кӏь

Other Languages: Some entries in the dataset include words originating from other languages, such as Russian or Turkish. These words have been phonetically transcribed using the Kabadrian alphabet, with characters chosen to approximate their original pronunciation as closely as possible.

Sample

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

1. Хьэщӏэ лъапӏэхэ!

2. Нэхъыбэ яхуэгъэхьэщӏэнукъым.

3. Гъуэгурыкӏуэм абы и нэм къыфӏэнэ ӏуащхьэхэр, псыежэххэр, жыгей мэзыжьхэр, къуакӏэбгыкӏэхэр фӏыуэ ицӏыху хуэдэу къыфӏэщӏырт.

4. Еплъыт, щӏыбым уэс къыщесырэ!

5. Guaşe mıxhu jant'ak'ueş ,

Sources

All sentences in this dataset were created specifically for the Common Voice project by native Kabardian speakers. The sentences were carefully crafted to cover a wide range of topics and linguistic structures, ensuring a diverse and representative corpus for speech recognition training.

Text domains

Processing

The text data in this dataset has undergone several processing steps to ensure quality and consistency:

Recommended post-processing

These sentences are provided as-is, and no additional post-processing is required. However, users may choose to perform Unicode normalization (NFC or NFD) depending on their specific use case.

Although the sentences (questions/translation texts) have been written and verified using correct Unicode characters via the ady & kbd keyboard for Windows, it is possible that some text inputs beyond our control may have been submitted due to Common Voice’s open participation model.

After the datasets are published, we re-audit these entries and take the necessary steps to ensure corrections in the next release.

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

• If you want to join kbd community, start here: https://bit.ly/cv_circassian_start_here

You can find more information about how to participate in the Common Voice Project on the following page:

• Common Voice translators on Pontoon

• Common Voice Communities

Discussions

Comming soon.

• Common Voice on Matrix

• Common Voice on Discourse

• Common Voice on Discord

• Common Voice on Telegram

Contribute

We do not advice to contribute to this dataset directly, because of writting system-keyboard issues. Please, read the section "Writing system" above and the following article: https://www.nemerko.org/the-importance-of-adyghe-keyboard-usage-for-artificial-intelligence-training/

If you want to contribute to the Kabardian language, please visit the following links: https://bit.ly/cv_circassian_start_here

A special thanks to all the volunteers who contributed to this dataset.

• Speak

• Write

• Listen

• Review

Acknowledgements

Datasheet authors

M.Ugur Nemlioglu
Murat Topçu

Murat Topçu Elizaveta Gogunokova Saida Abregova M.Uğur Nemlioğlu

Bülent Özden (Technical Advisor)

Funding

This dataset was partially funded by the Open Multilingual Speech Fund managed by Mozilla Common Voice.

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