Audio8-TTS-0.6B-MLX

A repack of Edge0/Audio8-TTS-Realistic-0.6b (source revision 1e857568bfd4f02931aca3988ee773a44d2d26d3) by Edge0, licensed under Apache 2.0. All credit for the model goes to Edge0.

This repo only rearranges the weights so the Fish DualAR loader in TinyTrashLabs/mlx-audio-swift (PR #20) can load them: the slow and fast transformer tensors are stored under model.* in model.safetensors, the codec is converted from codec.pth to codec.safetensors, and config.json is written in the fish_qwen3_omni layout. No weights are altered or quantised.

Audio8-specific config keys

max_frames_per_text_token: 12.0
reference_budget_scale: 1.95

Everything else in config.json is the Fish layout with scale_codebook_embeddings: false.

Licence

Apache 2.0. See LICENSE and NOTICE (copied from the source repo).

Downloads last month: 37

Safetensors

Model size

0.6B params

Tensor type

BF16

MLX

Hardware compatibility

Quantized

Model tree for tinytrashlabs/Audio8-TTS-0.6B-MLX

Base model

Edge0/Audio8-TTS-Preview-0.6b

Finetuned

Edge0/Audio8-TTS-Realistic-0.6b

Finetuned

(5)

this model