| license: other | |
| tags: [resonance, anima, tts, speech, from-scratch] | |
| # FatStinkyPanda · Resonance — step 38,000 | |
| A from-scratch **neural text-to-speech** engine (SNAC 24 kHz codec), trained in the **Anima** project (https://github.com/FatStinkyPanda/Anima). The published `resonance.safetensors` holds the trained tensors; the architecture loads them via the project's private engine. | |
| > **Proprietary architecture.** The architecture and training innovations are NOT public. This repo hosts the trained **weights + tokenizer only**; the config + modeling code stay private. Benchmarks run on **neutral free compute** and are published openly so the numbers are verifiable without exposing how the model works. | |
| ## Latest held-out benchmarks (neutral compute) | |
| - **behavior**: 1.0 | |
| - **layer_health**: 1.0 | |
| - **stability**: 1.0 | |
| - **coarse_accuracy**: 0.0618 | |
| - **overall**: 0.6506 | |
| - **tf_coarse_accuracy**: 0.0618 | |
| - **tf_loss**: 22.0305 | |
| - **tf_loss_coarse**: 8.1523 | |
| Created & released by **FatStinkyPanda**. For access or collaboration, contact the creator. | |