← Audio

CartesiaCartesia

Cartesia AI, Inc.

云端Cloud TTSTTS 语音agentVoice agent 多语言Multilingual API平台API platform 免费+订阅Free + subscription 商业Commercial partial lifecycle: active 核验Checked 2026-10-30

Cartesia 是「实时低延迟 TTS（Sonic）+ 转录模型（Ink）+ 托管语音 agent」三线平台 （Cartesia AI, Inc.，条款当日在册）：44 语言语音面，voice agent 底座场景（客服/外呼/招聘筛选）主打。

Cartesia is a three-line platform: realtime low-latency TTS (Sonic) + transcription model (Ink) + managed voice agents (Cartesia AI, Inc., on record same day): 44-language speech coverage, voice-agent infrastructure (support/outbound/recruiting) is the pitch.

适合与不适合When to use it, when not to

适合：低延迟实时 TTS（voice agent 底座）、44 语言覆盖、托管语音 agent 开箱。不适合：要大陆直连备案入口、要选第三方模型、要本机跑模型。

Fits: low-latency realtime TTS (voice-agent backbone), 44-language coverage, out-of-box managed voice agents. Not a fit: direct mainland access with a filing, picking third-party models, running models locally.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-30

出品属地Origin: 国外出品foreign-made　Cartesia AI, Inc.（条款页当日直抓在册主体；卡内基梅隆背景未逐条核验，按条款页在册口径写 foreign）。Cartesia AI, Inc. (entity on record in the terms page direct-fetched same day; CMU background not verified item by item, recorded as foreign per the terms-page evidence).
国内使用Direct access: 大陆可直连direct OK　cartesia.ai 与 /pricing、/legal/terms、docs.cartesia.ai 大陆直连实测 200（2026-10-30）。cartesia.ai and /pricing, /legal/terms, docs.cartesia.ai measured 200 direct from mainland (2026-10-30).
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-30 检索百炼公示页，未见 Cartesia 条目）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (Bailian disclosure page searched 2026-10-30; no Cartesia entry); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 自有模型线当日在册：Sonic（TTS）、Ink（speech to text/转录模型）+ Managed Agents（语音 agent 托管）； 44 语言语音面在册；不开放选第三方模型。
Its own model line on record same day: Sonic (TTS), Ink (speech-to-text model) + Managed Agents (hosted voice agents); a 44-language speech surface is on record; no third-party model picking.
运行位置Where it runs: 云端 API + 托管 agent（docs.cartesia.ai 当日在册）；官网招牌 Realtime AI models and voice agents—— 低延迟实时合成是主打；本机不跑模型。
Cloud API + hosted agents (docs.cartesia.ai on record same day); the site headline is "Realtime AI models and voice agents" — low-latency realtime synthesis is the pitch; no local model.
本地文件Local files: 未见官方声明的本机数据面；合成与转录结果在 Cartesia 云端账户侧，下载的音频文件归本机但会话记录与音色配置不落地。
No locally-held data surface declared by the vendor was found; synthesis and transcription results live on Cartesia's cloud account side — downloaded audio files are yours locally, but session records and voice configurations do not land on your machine.
关机后的任务Tasks after shutdown: API 形态（流式合成为主打叙事）；Managed Agents 是常驻托管形态；排队/续跑细节未逐条核验。
API form (streaming synthesis is the lead narrative); Managed Agents is an always-on hosted form; queueing/resume details not verified item by item.
工具与扩展Tools & extensions: 产品面当日在册：Sonic TTS、Ink STT、Voice Agents（客服/预约/外呼/SDR/招聘筛选场景页在册）； docs 站完整（get-started/changelog 2026 在册）。MCP 支持未核验到官方声明。
Product surfaces on record same day: Sonic TTS, Ink STT, Voice Agents (scenario pages for customer support / appointment scheduling / outbound calls / SDR / recruiting screening); the docs site is complete (get-started/changelog 2026 on record). No MCP support verified in official statements.
上下文与记忆Context & memory: Voice Agents 线有会话状态面（客服场景）；具体记忆机制未核验到官方声明（未知就说未知）。
The Voice Agents line has a session-state surface (support scenarios); the specific memory mechanism was not verified in official statements (say unknown when unknown).
权限与限制Permissions & limits: 账号体系（免费+订阅+企业 Contact Sales）；克隆音色授权面未逐条核验（docs 面未细读）。
Account system (free + subscription + enterprise Contact Sales); the cloned-voice consent surface not verified item by item (the docs surface not read in depth).
适合什么任务Best for: 适合：低延迟实时 TTS（voice agent 底座）、44 语言覆盖、托管语音 agent 开箱。不适合：要大陆直连备案入口、要选第三方模型、要本机跑模型。
Fits: low-latency realtime TTS (voice-agent backbone), 44-language coverage, out-of-box managed voice agents. Not a fit: a filed direct-mainland entry, picking third-party models, running models locally.

头号误解Top misconception

主打是「实时低延迟」——官方 latency 数字是自家口径（docs/官网在册），第三方未复核；选型对比时按同条件自测The pitch is “realtime low latency” — official latency numbers are the vendor's own claims (on record in docs/site), not third-party verified; self-test under equal conditions when comparing
Ink 是语音转文本模型（不是 TTS 附属功能）——把它当「TTS 厂商的转写副线」会低估其独立模型线定位Ink is a speech-to-text model (not a TTS side feature) — treating it as “the transcription side of a TTS vendor” undersells its independent model line
官网 changelog 页 404——现行 changelog 在 docs.cartesia.ai/changelog（2026 年在册），别按主域路径判更新面缺失The main-site changelog path 404s — the current changelog lives at docs.cartesia.ai/changelog (2026 on record); don't judge the update surface missing by the main-domain path

未知项清单Known unknowns

档位价格具体数字（pricing 页 JS 渲染，未逐条抓取）
克隆音色授权校验流程（docs 面未细读）
Sonic 模型架构细节（官方有 Research/News 页，未逐条核验）
MCP 支持（未核验到官方声明）

Exact tier prices (the pricing page is JS-rendered; not fetched item by item)
Cloned-voice consent verification flow (the docs surface not read in depth)
Sonic model architecture details (the official Research/News pages exist; not verified item by item)
MCP support (no official statement verified)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-30，内容更新日content updated on 2026-10-30。

类型Type	名称Name	链接Link
official	Cartesia · 官网（Sonic/Ink/Managed Agents/44 语言产品线，2026-10-30 直抓 200）Cartesia · official site (Sonic/Ink/Managed Agents/44-language product lines, direct-fetched 200 on 2026-10-30)	https://cartesia.ai
official	Cartesia · Terms of Service（主体 Cartesia AI, Inc.，2026-10-30 直抓 200）Cartesia · Terms of Service (entity Cartesia AI, Inc., direct-fetched 200 on 2026-10-30)	https://cartesia.ai/legal/terms
docs	Cartesia · docs（get-started 概览，2026-10-30 直抓 200）Cartesia · docs (get-started overview, direct-fetched 200 on 2026-10-30)	https://docs.cartesia.ai
changelog	Cartesia · docs changelog（2026 年更新线，2026-10-30 直抓 200）Cartesia · docs changelog (2026 update line, direct-fetched 200 on 2026-10-30)	https://docs.cartesia.ai/changelog
official	Cartesia · Pricing（档位面，2026-10-30 直抓 200）Cartesia · Pricing (tier surface, direct-fetched 200 on 2026-10-30)	https://cartesia.ai/pricing

实测记录Measured runs

本档未跑音频质量与延迟测试；结论只覆盖能力面与许可面。

实测建议（三条，都指向官方可读页没给的东西）：

官方 latency 口径下的端到端首包延迟（官方自称 realtime，第三方未复核；与 ElevenLabs/Murf 同条件对测最有信息量）
Ink 转录的中文覆盖与质量（44 语言是语音面总口径，Ink 单独语言面未逐条核验）
Managed Agents 的通话成本结构（分钟计费细节未逐条核验）

This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to audio products).

Suggested measurements (three, all pointing at things the official readable pages do not give):

End-to-end first-packet latency under the official latency claims (the official line self-claims realtime; no third-party re-check; head-to-head against ElevenLabs/Murf under equal conditions is the most informative)
Ink's Chinese coverage and quality (44 languages is the aggregate speech figure; Ink's per-language surface not verified item by item)
Call cost structure of Managed Agents (per-minute billing details not verified item by item)