Sinoosoida/SpeechRu

Russian Podcasts (unlabeled)

~186k unlabeled Russian-language podcast episodes scraped from the web, packaged as Parquet shards with the audio bytes embedded. The audio has no transcripts — this is an unsupervised / self-supervised audio corpus, suitable for ASR pre-training, speech-representation learning, TTS data mining, audio classification, and similar tasks.

Each row contains:

• —audio — the podcast episode (MP3, mostly 128 kbps / 44.1 kHz stereo), decoded on-the-fly via the Audio feature.
• —a set of flat scalar metadata columns: id, name, lang, duration_sec, audio_bytes, rating, rating_votes, rating_scores, read_count, reviews_count, citations_count, author / narrator names, publisher_name, genres_names, tags_csv, series_name, written_dt, updated_at, annotation_plain, annotation_html, and a few more.
• —raw_metadata_json — the full original metadata record serialized as a string.

Loading examples

from datasets import load_dataset

# stream the whole thing
ds = load_dataset("Sinoosoida/SpeechRu", split="train", streaming=True)
for ex in ds.take(3):
    print(ex["name"], ex["duration_sec"], ex["rating"])
    audio = ex["audio"]  # AudioDecoder

# metadata-only: skip downloading audio bytes
ds = load_dataset(
    "Sinoosoida/SpeechRus", split="train", streaming=True,
    columns=["id", "name", "duration_sec", "rating", "genres_names"],
)

# filter by metadata
ds = load_dataset(
    "Sinoosoida/SpeechRus", split="train", streaming=True,
    filters=[("duration_sec", ">", 3600)],  # episodes longer than 1 hour
)

The audio was scraped from publicly reachable sources on the internet and is provided as-is for research use.