moss-tts

Audiobook creation app supporting too many TTS models (Qwen3-TTS, OmniVoice, VibeVoice, etc), focused on high-quality output. Plus audio-synced reader web app and standalone server component.

CPU-only inference pipeline for MOSS-TTS 8.4B in PyTorch, forcing Float32 to bypass BFloat16 issues and adding selective quantization on the LM backbone. Runs efficient text-to-speech on hardware without a GPU.

The best voice-cloning TTS on the planet. Clone any voice, 31 languages, runs 100% offline. Free forever.

Roo Voice — local, reference-conditioned single-voice TTS. One-click app + models for Apple MLX, NVIDIA CUDA, AMD Vulkan/ROCm, and NPU (ONNX).

🔊 Local cloned-voice feedback for Claude Code + Codex on Apple Silicon — MOSS-TTS-Nano on MLX, 0.3s first sound, spoken summaries, multi-session queue, offline after download.

Reproducible deployment of OpenMOSS-Team/MOSS-TTS-v1.5 on Kaggle GPU T4 x2 using the original BF16 safetensors checkpoint—no GGUF, no quantization, no model-weight rewriting. Includes explicit two-GPU module placement, staged backbone/codec execution, fail-closed verification, multilingual qualification, benchmarks, and release evidence.

Local text-to-speech app for Home Assistant

Auditable script-to-video workflow with MOSS-TTS, image keyframes and HyperFrames

Empirical evaluation and latency benchmark suite for OpenMOSS MOSS-TTS (Realtime & Local-Transformer) focusing on German speech synthesis.