Tag
Fish Audio 发布了 Drama 3 模型的权重，支持用自然语言指导台词演绎并可在句中切换表演方式，引发关于开源语音模型是否将迎来“DeepSeek 时刻”以及 ElevenLabs 等闭源模型护城河的讨论。
Openspline 是一个开源的实时数字人头像层，可通过 5 行代码接入现有语音 AI 智能体，在本地 GPU 上生成口型同步的实时视频，RTX4090 上可达 90+ FPS。
作者以语音 AI 编排平台产品负责人的视角，分析 Jev 及同类分类模型在电话语音机器人中的实际用途，并梳理了 Laya、Clef、CMF、Nimble、Kev 等竞品生态。文章指出单一 LLM 无法独立支撑生产级对话，分类模型在延迟、成本和结构化输出上仍不可或缺。
Bland 推出 iMessage 智能体功能，可在30秒内将任意商业网站转化为带品牌标识的短信客服线路，支持实时抓取价格、库存并接受支付。
Grok Voice 3.0 发布，被称为全球最智能的语音到语音模型，响应时间低于1秒，面向客服通话场景且成本低于其他前沿模型。
Grok Voice 3.0 tops Artificial Analysis's Speech-to-Speech Index with a score of 86.9, beating OpenAI and Google, while powering hands-free voice features like calendar checks, email replies and Grok Bot tasks inside Tesla vehicles handling millions of daily conversations.
该论文提出 PeopleMakingConversation 基准，通过 176 段非脚本对话中的 427 个停顿点和 482 个话轮结束点，评估六个实时语音系统在何时开口、延迟、回复长度与对话记忆等方面的自然对话表现，并公开数据与评测框架。
多位语音AI公司高管在HumanX大会上表示，尽管全双工模型和语音AI投资热潮不断升温，语音AI仍未迎来自己的“ChatGPT时刻”，低延迟推理、ASR准确率和对话透明度仍是关键挑战。
一位 Shopify 商家称赞 @bot 仅用一天就为其店铺搭建了基于 xAI 语音的 AI 电话客服，可查询订单、修改订单并转接人工。
ElevenLabs发布了用于ElevenAgents的合成人声检测功能，并加入Personal Agent Protocol，以提升AI代理间通信的安全性与透明度，应对日益增多的AI代理拨打商业电话的趋势。
VOYGR (YC W26) ran an SF Tech Week experiment calling 1,000+ San Francisco businesses with an AI phone agent via their PlaceCall API, finding that about 2 in 3 businesses hung up when told they were speaking with an AI, with vets and dentists most receptive and bars and dry cleaners least.
Decagon 推出其新型语音模型 Chord，由 Decagon Labs 团队开发，在 Modal 上完成训练与部署，具备低首帧音频延迟和高并发下的稳定表现。Chord 经过后训练以保持语速和语音自然度，并在所有部署中提升了问题解决率。
The author praises 'realtime Talk' as a superior voice interaction solution, claiming it delivers real-time, interruptible voice experiences rivaling commercial ChatGPT and Gemini apps while leveraging the user's own LLM agent with full context and tools.
ClawCall 是一款让 ChatGPT、Claude、Grok 等 AI 代理代为拨打电话的产品，可自动处理语音菜单、排队等待、与人工对话并汇报结果，支持需要决策时转接用户本人。产品在 Product Hunt 上线，免费 30 次通话，之后每月 4.99 美元。
A user praises Grok's new voice mode in the bot, calling it incredibly smooth and able to actually perform real tasks, with a demo video showcasing audio-enabled voice call interactions.
Shubham Saboo announced a free 30-minute masterclass teaching AI PMs how to build a multimodal voice AI agent that can see, think, talk, and draw, covering the full workflow from problem to idea to spec to a working agent.
An opinion arguing that AI assistants like Alexa and Siri don't need to mimic human voices — stilted robot speech is fine as long as users know they're talking to an AI, raising the question of transparent machine identity over human-like realism.
Podonos is launching Onepin, a production-ready voice AI tool that checks each generated line for naturalness and pronunciation, correcting mispronunciations like brand names automatically so a finished voiceover comes out directly from a script.
Onepin launches as a model routing and quality-control layer for voice generation, sitting on top of existing TTS providers to normalize scripts, select voices per line, and automatically detect and fix pronunciation or naturalness issues at scale.
UTTER IN is a voice-powered productivity app that turns a single voice note containing multiple requests in English or Arabic dialects into separate reminder cards, calendar events, and saved places, which the user must approve before syncing to Google, Outlook, or Apple Calendar.