Common Voice v24 English - en-AU subset for Everything Open 2026

Tutorial information

• Everything Open: https://2026.everythingopen.au

• Tutorial overview: https://2026.everythingopen.au/schedule/presentation/6/

• Tutorial GitHub repo: https://github.com/Mozilla-Data-Collective/tutorial-whisper-fine-tuning-australian-EO2026

Preprocessing information

This dataset was extracted from Common Voice v24 English by filtering on the accent field, after assessing the Australian-related accents in the dataset.

The duration of each clip was also calculated, to assist in identifying very long or short clips, and this is stored in ms in the field duration_ms.

File structure

• audios => contains the audio files in the format id.mp3 where id is the unique identifier of the clip.

• commonvoice-v24_en-AU.csv a CSV-formatted file.

The CSV fields are:

• original row ID from Common Voice v24 English

• client_id: unique identifier for each speaker

• path: the filename of the audio file

• sentence_id: a unique identifier for each written sentence

• sentence_domain: a string description of the topic domain of the sentence (may be null)

• up_votes: integer indicating how many up votes this clip has

• down_votes: integer indicating how many down votes this clip has, allows for exclusion

• age: age range of speaker, if provided (may be null)

• gender: gender identify of speaker, if provided (may be null)

• accents: accent descriptor

• locale: ISO-639 locale (all samples in this dataset are en)

• segment: not applicable to this dataset, included to provide interoperability

• duration_ms: duration in milliseconds of the audio file, calculated using librosa

Composition

This dataset comprises 55673 rows of Australian-accented elicited (read) English speech.

The total length of time is approximately 4.68 minutes.

Accents represented

• Australian English

• General Australian

• South Australia

• Educated Australian Accent

• Sydney - middle eastern seaboard Australian

• Queenslandish