Datasetpublic

Peacockery/farsi-asr-corpus-v4

farsi-asr-corpus-v4 985 hours of Farsi ASR training data across seven corpora: Common Voice 25 (324 h), Thomcles Farsi speech (298 h), Farsi YouTube (205 h), Mana TTS (97 h), Neyshekar (37 h), FLEURS fa_ir (13 h), WorldSpeech (11 h). This is the dataset behind Peacockery/omni-ctc-300m-farsi (8.5% WER on FLEURS test). Layout Hive-partitioned parquet under version=0/corpus=<source>/split=<split>/language=fas_Arab/. Each row holds text (the normalized label)… See the full description on the dataset page: https://huggingface.co/datasets/Peacockery/farsi-asr-corpus-v4.

sourceHugging Facecc-by-4.0updated 4mo agoView on Hugging Face

0likes49downloads

Dataset Card

farsi-asr-corpus-v4

985 hours of Farsi ASR training data across seven corpora: Common Voice 25 (324 h), Thomcles Farsi speech (298 h), Farsi YouTube (205 h), Mana TTS (97 h), Neyshekar (37 h), FLEURS fa_ir (13 h), WorldSpeech (11 h). This is the dataset behind Peacockery/omni-ctc-300m-farsi (8.5% WER on FLEURS test).

Layout

Hive-partitioned parquet under version=0/corpus=<source>/split=<split>/language=fas_Arab/. Each row holds text (the normalized label), audio_bytes (16 kHz mono FLAC as an int8 list), and audio_size (sample count). language_distribution_0.tsv lists hours per corpus.

How the labels were made

Scripted corpora (Common Voice, FLEURS, Mana TTS) keep their original transcripts. Web-scraped corpora (YouTube, Thomcles, WorldSpeech, Neyshekar) carry machine labels verified against an independent ElevenLabs Scribe pass and filtered on agreement; this is the fourth iteration of that verification ("scribe v4"). Text is normalized for CTC training (hazm orthography, ZWNJ handling, numbers expanded to Farsi words).

Machine-labeled portions are unverified by native speakers.

Sources

FLEURS: CC-BY-4.0, google/fleurs. Common Voice: CC0, Mozilla. Remaining corpora: public web sources, labels produced by this project.