THE CURIOUS SIDE OF AI
TTS / DATASET GUIDE
LJ Speech 1.1
The original publisher provides 13,100 clips from one speaker, approximately 24 hours. Use the official source, not an unverified mirror.
Source reviewed 2026-10-24 · AiBrainWorX editorial guide
- Purpose
- Single-speaker TTS preparation
- Format
- 22,050 Hz WAV + pipe-delimited transcript metadata
- Workflow
- External Qwen TTS single-speaker SFT
- License
- Public domain (publisher declaration)
01 / COMPATIBILITY
Resampling and audio/text/ref_audio JSONL preparation required
The Qwen TTS Base recipe is a separate audio workflow. Confirm voice-use rights, prepare/resample audio and create the required reference-audio fields.
02 / PREPARE LOCALLY
Keep the first run small
- Review the selected source subset, license and data rights.
- Prepare the format required by the chosen trainer. Audio, image classification and text SFT are different workflows.
- Set aside evaluation examples and inspect the final records before training.
- For the launcher text worker, choose a local JSONL file and supported HF checkpoint, then run Preflight.
Our test status
Source reviewed only; not downloaded or trained.
03 / ORIGINAL SOURCE
Review files and terms
This is an original concise guide, not a mirror or relicense of the dataset.
Open original publisher ↗Reviewed source revision: Publisher page reviewed 2026-10-24