tts

pyVideoTrans is an open-source video translation tool that supports automatic speech recognition, subtitle translation, AI dubbing, and video synthesis. It integrates multiple ASR, translation, and TTS engines, making it suitable for cross-language video production and localization.

MOSS TTS 1.5 is a new text-to-speech model with voice cloning capabilities, offered via a Hugging Face Space, and is considered better than Fish Audio S2 Pro due to open licensing.

NVIDIA released an optimized ONNX version of the Kokoro TTS model on Hugging Face. The 82M parameter model is lightweight, fast, and ready for commercial use.

pibot is now fully local, using Parakeet for STT, Qwen3-tts for TTS, and Qwen 3.6 as the local multimodal LLM via llama.cpp, with Rust/mlx-c based inference engines, achieving zero Python dependencies.

seshat-tts is an open-source tool that enables real-time game narration with voice cloning, using OCR or an LLM for text extraction and local synthesis with pocket-tts. Voice cloning takes ~10 seconds on an RTX 2070 Super and runs on CPU after caching.

A free, open-source AI skill that generates fully animated and narrated explainer videos locally using Claude for animation code, Edge TTS for voice, and ffmpeg for rendering—no subscriptions or API keys required.

MOSS-TTS v1.5 is an updated open-source text-to-speech model with improved multilingual synthesis (supporting 31 languages), more stable zero-shot voice cloning, and explicit inline pause control.

A user-created benchmark for comparing local TTS tools, with results for Windows and Mac, and Linux testing pending. Includes an HTML results page and GitHub repository.

Cartesia launches Sonic 3.5, a new state-of-the-art TTS model supporting 42 languages, taking the #1 spot on the Artificial Analysis Speech Arena Leaderboard.

Supertonic 3 is a 99M parameter open-source TTS model that runs entirely on-device, beating ElevenLabs on a Raspberry Pi with 167x faster than real-time performance on a laptop CPU.

StepAudio 2.5 is a unified audio-language model that achieves state-of-the-art results across ASR, TTS, and real-time spoken interaction by leveraging task-tailored reinforcement learning from human feedback to optimize shared representations.

StepFun launches Step Plan subscription at $6.99/month, integrating LLM, TTS, ASR, image generation, and other AI models. Supports direct OpenAI SDK connection, applicable for voice cloning, meeting transcription, AI podcast generation, etc.

A user benchmarks 21 consumer GPUs on vast.ai running a small TTS model (OmniVoice) with peak VRAM of 5GB, comparing performance relative to real-time and to an RTX 3090.

Audiblez is an open-source tool that quickly converts EPUB ebooks into high-quality M4B audiobooks. It uses the Kokoro-82M voice model, supports multiple languages and a graphical interface, and can be installed with a single pip command.

Hermes Web UI v0.5.25 released. New features: group chat room reset/clone, configurable state directory, MiMo TTS (supports Mandarin, Taiwanese, Cantonese, English), and improved Provider model list proxying via backend.

OpenMOSS is a standalone C++/GGML port of MOSS-TTS that provides a self-contained binary for text-to-speech and voice cloning, leveraging a Qwen3-8B backbone and 32 RVQ audio codebooks. It includes a CLI for one-shot synthesis and an HTTP server for repeated generations.

Supertone released Supertonic 3, an open-source TTS model with 99M parameters that runs faster on CPU than a 2B model on A100, supporting 31 languages and ONNX Runtime for fully local inference.

Supertonic is a new open-source TTS engine that runs on-device via ONNX, supporting 31 languages and outperforming ElevenLabs in speed, even on a Raspberry Pi without a GPU.

Violin is an open-source end-to-end video translation and video Q&A tool, integrating ASR, LLM translation, and TTS. It supports style adjustment and content re-creation, and can answer questions about video content.

After 6 months running a voice AI agent for service businesses, the author reveals that real-world latency is bimodal (median ~800ms, p95 ~2.4s) and this p95 determines user perception. Issues like VAD misfires, function call degradation with long prompts, and TTS quality matter more than LLM choice, with multilingual support adding significant costs.