SOTA-Class TTS at Compact Scale
Installation and inference
All releases · 中文 · Deployment · SFT
Installation
Run from the repository root. Python 3.10+ and a CUDA GPU are recommended for 0.6B; the 0.1B model card recommends Python 3.11+.
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
Choose a release
Choose one model ID below and set MODEL. Every following command uses that release. A complete local checkpoint directory is also accepted.
| Release | MODEL |
|---|---|
| Realistic Air | Edge0/Audio8-TTS-Realistic-Air |
| Preview 0.6B | Edge0/Audio8-TTS-Preview-0.6b |
| Preview 0.1B | Edge0/Audio8-TTS-Preview-0.1b |
export MODEL=Edge0/Audio8-TTS-Realistic-Air
# Preview 0.6B: export MODEL=Edge0/Audio8-TTS-Preview-0.6b
# Preview 0.1B: export MODEL=Edge0/Audio8-TTS-Preview-0.1b
If you arrived from a Preview guide, select its Preview ID. Without --model, the scripts still default to model/audio8_tts_0_6B_preview/; this guide always passes the model explicitly.
Inference
For best synthesis quality, keep each input within 150 characters. Longer text may reduce generation quality; split it into shorter segments when needed.
Zero-shot voice cloning
Use your own recording at reference.wav. Its transcript should match the spoken content exactly.
python audio8_tts_infer.py \
--model "${MODEL}" \
--text "Welcome to audio8_tts." \
--reference-audio reference.wav \
--reference-text "Transcript of the reference recording." \
--output outputs/clone.wav
Generation without a reference
python audio8_tts_infer.py \
--model "${MODEL}" \
--text "This utterance does not use a reference voice." \
--output outputs/no_reference.wav
Batch inference
Each line in the input manifest is an independent JSON object. Relative audio paths are resolved from the manifest directory.
{"id":"sample_001","text":"Target text","reference_audio":"audio/ref.wav","reference_text":"Reference transcript"}
{"id":"sample_002","text":"Text without a reference voice"}
python audio8_tts_infer.py \
--model "${MODEL}" \
--input-jsonl data/prompts.jsonl \
--output-dir outputs/batch \
--batch-size 2
The batch command writes manifest.jsonl and failures.jsonl. Existing WAV
files are skipped unless --overwrite is passed. See
python audio8_tts_infer.py --help for sampling and code-saving options.