Kokoro-82M
by HexgradThe two-paragraph Gettysburg excerpt, spoken with Kokoro-82M on Apple Silicon (MLX) or PyTorch (~23s).
Examples & Models
Kokoro, Qwen, F5, Higgs, and Chatterbox are the supported speech backends. Each model narrates the identical two-paragraph Gettysburg excerpt so you can directly compare audio timbre, pacing, pronunciation, and latency.
Speech Model Comparison
Compare five open model architectures running on Apple Silicon (MLX) and AWS Batch:
82M params
Qwen3-TTS
by Alibaba QwenThe same Gettysburg excerpt, spoken with Qwen3-TTS 0.6B CustomVoice with the Ryan preset (~48s).
F5-TTS
by SWividThe same Gettysburg excerpt, spoken with non-autoregressive F5-TTS via MLX on Apple Silicon.
Higgs Audio v3
by Boson AIThe same Gettysburg excerpt, spoken with Boson AI's expressive 4B Higgs Audio v3 via MLX on Apple Silicon.
Chatterbox-TTS
by Resemble AIThe same Gettysburg excerpt, spoken with Resemble AI's hybrid LLaMA-520M and Matcha-TTS flow matching.
Fish-Speech
by Fish AudioThe same Gettysburg excerpt, spoken with Fish Audio's dual-autoregressive model with natural prosody and breathing.
Embed Features & Theming
Customizing player appearance and article content extraction:
- Themed PlayerCSS variablesThe same Gettysburg excerpt with Kokoro, styled via CSS custom properties (--auritus-*).
- Ignore & Pronounce MarkupArticle parsingJohn Stuart Mill's On Liberty demonstrating data-auritus-ignore to skip ads and data-auritus-pronounce.