Tag
Silencio-2 is an open-weight 18M-parameter TTS fine-tune of Kyutai's Pocket TTS that follows inline emotion and sound tags (laugh, sigh, whisper, angry, etc.) and runs ~20x realtime on CPU. The author shares training details and lessons, notably that an AST/AudioSet clip-level filter silently dropped rare sound events, so it was removed entirely.
A new benchmark called Turing TTS measures how often human listeners mistake AI-generated voices for real human speech, with Gemini 3.8 Flash TTS currently the best at fooling people in A/B voice identification tests.
The next breakthrough in voice AI may come from rethinking how we evaluate models rather than crowning a new top performer, with Voice Arena's Turing TTS benchmark measuring how often human listeners mistake AI voices for real ones.
Introducing Turing TTS, an evaluation measuring how often human listeners mistake AI-generated speech for real human speech; even the best TTS models fooled listeners only ~10% of the time, prompting discussion on better metrics for voice AI progress.
为 audio.cpp 新增 any-tts 配方，将 Kokoro TTS 与 OpenVoice 的 ToneColorConverter 结合，仅需约 600 MB 额外 VRAM 即可实现免训练的语音克隆能力。
Deepgram's new Flux TTS targets a critical accuracy bottleneck in voice-agent pipelines, achieving a 3.4% word error rate on tricky alphanumeric strings like medication names and tracking numbers while preserving natural-sounding speech through the full voice loop.
Armin Ronacher explores custom voice creation with Qwen3-TTS, covering its Base, CustomVoice, and VoiceDesign variants and showing how to blend voice embeddings with short reference audio to build a unique narrator for a dungeon-crawler game project with his kids.
A free in-browser TTS tool that runs entirely on-device using a 2.7 MB model on a custom 136 kB WebAssembly engine, making Kokoro-distilled Paradee 12.9x faster while maintaining quality. The author breaks down each optimization step and lets users benchmark it live in their browser.
Gradium 发布了首个开源模型 Phonon，这是一组 31M 和 40M 参数、专为边缘/设备端运行设计的 TTS 模型，其推理栈完全开源并支持 CPU、GPU、NPU 等 10 多种后端（大多基于 Rust）。
Kyutai announces that its Pocket TTS model, which supports voice cloning, has been shrunk to just 1.9 million parameters—roughly 10x smaller than before—demonstrating that high-quality speech synthesis can run on extremely lightweight models.
VoiceManager 是一个低延迟的流式语音到语音（STT/TTS）本地服务器，通过单一 WebSocket 进行流式传输，支持带打断（barge-in）的多轮对话；HermesConsole 则是一个 LCARS 风格的 Kotlin 客户端参考实现，展示唤醒词、流式语音回复、工具调用及会话上下文保持等功能，采用 MIT 协议开源。
inclusionAI 发布 Ming-3.0-audio-flash，一个 126B 总参数（每步激活 6.5B）的原生音频-语言 MoE 模型，支持中英文语音识别、语音问答、长音频理解、音频推理以及可选的语音合成输出，采用 MIT 许可证。
A developer built Red Curtain to run Fish Audio's Drama 3 text-to-speech model (35B MoE, ~3B active) on a 16 GB Radeon RX 9070 despite its official 48 GB NVIDIA requirement, offloading FP8 experts to system RAM and caching active ones on the GPU. The tool ships with a web UI supporting voice cloning and multi-speaker scripts, and the author shares tuning tips and a known first-line silence issue.
A team shares how they built a production voice-agent stack using self-hosted Pipecat, Mai Transcribe 2, a Sarvam conversational LLM, and a Malayalam fine-tuned VoxCPM2 TTS, achieving 20 concurrent low-latency voice streams on a single L4 GPU with an automated WhatsApp post-call handoff.
ELARSpeech introduces a 30,202-hour speech dataset covering 675 underrepresented languages curated from the Endangered Languages Archive, along with trained ASR/TTS models and a full data pipeline to advance long-tail multilingual speech technologies.
一位开发者开源了本地运行的影视配音工具 dub-anything，利用 OmniVoice、IndexTTS、Chatterbox 和 Whisper 克隆并保留每位演员的原声进行多语言配音，让此前未被翻译的内容变得可看。
@no_stp_on_snek 发布了 Fish Audio Drama-3 的两个实验性量化版本（C0 和 C24B），通过自研 TurboQuant llama.cpp fork 在更低显存预算下运行语音合成模型，C24B 已在受限的 RTX 5090 环境下通过短语音测试。
Breeze-TTS-2 被认为是目前最强的本地 TTS 模型，但采用非商用许可；对于可免费商用的场景，Moss TTS 1.5 是当前最佳选择。
开发者发布 Amisoria——一款 iPhone 应用，可为运行在自家 Mac 上的 OpenClaw 网关智能体添加 3D 角色形象、神经语音、口型同步、免提对话与插话功能，支持 20 种语言，基础功能免费且部分开源。
Step-by-step tutorial: Use the AI agent Muse to run a Douyin account fully automatically. You only need to spend 10 minutes signing up and scanning a QR code; over the next 30 days, AI handles the entire workflow, from topic selection and scriptwriting to video generation, voiceovers (integrated with Doubao TTS), and publishing through browser automation.