Common Voice 25 Persian Raw: 428 Hours of Crowdsourced Speech

A 2026 release of the Mozilla Common Voice dataset for Persian (Farsi). It contains 390,953 audio clips representing 428.07 hours of recorded speech from 4,655 speakers, based on a text corpus of 341,251 sentences. The dataset was uploaded by Reza2kn.

Use Cases

• Training automatic speech recognition (ASR) models based on the large volume of validated Persian audio clips.
• Developing accent detection or speaker diarization tools based on the demographic and accent information mentioned.
• Building text-to-speech (TTS) systems using the aligned text corpus of over 340,000 sentences.
• Conducting linguistic research on Persian phonetics and speech patterns using the crowdsourced speaker data.

Strengths

• Contains 390,953 audio clips totaling 428.07 hours of speech.
• Includes data from 4,655 unique speakers, suggesting speaker diversity.
• 372.94 hours of the audio are validated for quality assurance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The description notes that only 22.8% of clips and 2.4% of speakers have an accent tag, indicating incomplete demographic coverage.

Provenance

Source

Mozilla Common Voice

Collection Method

Crowdsourced recordings from a text corpus.

Freshness

Last updated 2026-05-10 04:21:12.