Single-speaker synthetic Tajik speech, generated 2026-09-22..24. Sources: khazina (book readings), law (legal texts), books (book texts). CC-BY-NC-4.0. Research use only. Generated with an OmniVoice LoRA (omnivoice-lora-sl-panther2); OmniVoice is CC-BY-NC, so this corpus inherits the non-commercial restriction. Every manifest row carries licensenote: "CC-BY-NC-derived, research-only, never Sado product". This data must not be used in a commercial product. Manifest fields: clipid, text, wav, duration, refvoice, generator, synthetic, rmsdbfs, normalizationtargetdbfs, peakceilingdbfs, normalizationmethod, licensenote, sourcerow. All audio is loudness-normalised to -16 dBFS RMS with a -1 dBFS…
Independent publisher
Tohir Saidzoda
Tohirju
Models in Library0
Datasets in Library2
Models on Hugging Face135
Followers10
Compare the published evidence
This profile connects the 2 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, Text to speech.
Reported licenses: cc-by-nc-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
Uzbek audiobook recordings with machine-generated, structurally filtered transcripts. - audio/ — the raw full-length recordings, unmodified. - manifest/gold.jsonl — segments that pass every structural check. Training set. - manifest/review.jsonl — segments held back with the reason recorded. Kept, not discarded. - manifest/reject.jsonl — segments failing a hard check. - manifest/SUMMARY.json — counts, hours, and the gate thresholds actually used. audiohf is a path inside this repo, so a segment can be cut directly from audio/. text is the transcript as decoded; textnorm is case-folded and punctuation-stripped with the okina (U+02BB) preserved as a letter and apostrophe variants folded onto…