This public dataset is a reproducibility-oriented archive of multilingual audio tracks collected from publicly accessible institutional media pages. Files are grouped by source and published as source-level archives. The accompanying manifests preserve source URLs, language labels, extraction status, and verification metadata. The archive is intended for research and engineering evaluation; downstream users must respect the terms, licenses, privacy rules, and access policies of each upstream provider. Please cite the upstream source when redistributing or using a recording, and do not imply endorsement by the institutions represented in the source material.
Organization
HySI Lab
hysi-lab
Models in Library0
Datasets in Library3
Models on Hugging Face1
Followers5
Compare the published evidence
This profile connects the 3 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Speech recognition.
Reported licenses: cc-by-4.0, other. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Datasets
This hysi-lab mirror preserves NVIDIA's pinned HiFiTTS-2 metadata and the original LibriVox chapter audio referenced by its 22 kHz inventory. The 44 kHz metadata is also retained; it describes a subset and must not be counted as additional hours. Sources: https://huggingface.co/datasets/nvidia/hifitts-2 and https://librivox.org/. Please credit Ryan Langman et al., HiFiTTS-2: A Large-Scale High Bandwidth Speech Dataset (Interspeech 2025), NVIDIA, and the original LibriVox readers/books. Audio acquisition is in progress. Tar shards under archives/ (including hash subdirectories) contain original compressed chapter bytes at audio/.media and MANIFEST.jsonl, which maps each member to its…
Newly acquired English educational broadcast archives, with original publisher MP3 bytes, derived 16 kHz mono PCM WAV clips, automatic transcripts, word timestamps and per-episode speaker IDs. Collection was initialized on 2026-10-02; source publication dates are preserved separately. The initial pilot recordings were published in 2025; the historical expansion includes older source publication dates. This is an independent corpus made using the processing approach in mit-ccc/audio-scraper, from the group behind RadioTalk. It does not contain recovered original 2018–2019 RadioTalk recordings, and is not an exact replication of its Kaldi-era system, station sampling or talk-radio…