Audio8-TTS-Preview-0.6b-GGUF

GGUF conversion of Audio8/Audio8-TTS-Preview-0.6b for audio.cpp. No new training.

Original checkpoint: 0.6B DualAR — Slow 24×896 (14 heads, 2 KV), Fast 4×896, 10×4096 codebooks, 44.1 kHz codec (2048 samples/frame), 2048 context. Languages: yue/zh/nl/en/fr/de/it/ja/ko/pl/es/auto. See original model card for details and evaluation.

File

File	Tensors	Precision	Size
audio8-tts-preview-0.6b-q8_0.gguf	681 (226 AR + 455 codec)	q8_0	~1.4 GiB

Both embed config.json, tokenizer.json, tokenizer_config.json and model_specs/audio8_tts.json. q8_0 is the recommended default.

Conversion

• Source: model.safetensors + codec.safetensors from Audio8/Audio8-TTS-Preview-0.6b
• Tool: convert_audio8_tts.py + audiocpp_gguf (--family audio8_tts --model-spec model_specs/audio8_tts.json --type q8_0/bf16)
• Change: storage type/quantization only (q8_0). No architecture or training changes.

python tools/community_models/convert_audio8_tts.py \
  --model-dir /path/to/Audio8-TTS-Preview-0.6b \
  --converter build/bin/audiocpp_gguf \
  --type q8_0

Usage (audio.cpp)

# offline TTS
audiocpp_cli --task tts --family audio8_tts \
  --model audio8-tts-preview-0.6b-q8_0.gguf \
  --text "The quick brown fox jumps over the lazy dog." --out out.wav

# streaming + voice cloning
audiocpp_cli --task tts --mode streaming --family audio8_tts \
  --model audio8-tts-preview-0.6b-q8_0.gguf \
  --text "Hello world" --voice-ref ref.wav --reference-text "transcript of ref.wav" --out out.wav

# server (OpenAI-compatible)
audiocpp_server --config server.json
# POST /v1/audio/speech {"model":"audio8-tts","input":"...","stream":true,"stream_format":"sse","response_format":"pcm"}

Params: max_tokens (default 1024), temperature 0.7, top_p 0.9, top_k 50, seed.

License and Attribution

License: Apache-2.0 — same as upstream.

Upstream: Code and model weights are released under the Apache License 2.0. See NOTICE.

This repository is a Derivative Work (GGUF format conversion) under Apache-2.0 §1.

• Copyright retained by Audio8 — see upstream LICENSE/NOTICE (includes DualAR attribution to Fish Audio).
• This distribution includes Apache-2.0 LICENSE.
• Not affiliated or endorsed by Audio8.

Acknowledgements

Audio8 for Audio8-TTS-Preview-0.6b (Apache-2.0) and Fish Audio for DualAR architecture.

Limitations

Preview checkpoint; noisy/long/mismatched reference clips reduce stability. Obtain consent before cloning a voice and disclose synthetic audio.