Hojo-TTS-Light: open-source 40M/80M TTS benchmarked on a 2-core CPU, no GPU
HeyZoyaKhan · x · 2026-10-28
HojoAI's open-source (Apache-2.0, 613 stars) lightweight TTS series Hojo-TTS-Light ships 40M and 80M token-LM models supporting Chinese and English with average DNSMOS>4.0 on Seed-TTS eval. The 40M has 15 built-in voices; the 80M adds voice cloning from a few seconds of audio, both with optimized ONNX packaging.
A community Colab test on a 2-core CPU, no GPU, measured 1.7s per second of audio for 40M and 3.5s for 80M—neither real-time. Key deployment trade-off: keeping a model loaded avoids cold starts but occupies memory, and CPU inference support doesn't mean every CPU meets interactive latency needs.
More from Infra
- Mixed NVIDIA+AMD multi-GPU LLM setup benchmarks slower than a single card — LMasterGame · 2026-10-28
- 8 open-source repos to self-host a lean AI agent stack instead of stacking $20/month SaaS — SaiTarunRonanki · 2026-10-28
- Corning Optical Q3: sales up 44% to $2.38B, Gen AI revenue doubles y/y — demian_ai · 2026-10-28
- AI-powered formal verification finds real Redis/Valkey bugs, incl. a 23-char freeze — LingmingZhang · 2026-10-28
- 112 TPS on a single RTX 3090: LlamAmpere v0.6 pushes 27B models with 400K context — Brief-Tap-6616 · 2026-10-28
- Your coding agent burns up to 12k tokens before you type: 30 harnesses span 50x — yb2698 · 2026-10-28