跳到主要内容Skip to main content
← Audio

Resemble AIResemble AI

Resemble AI
云端Cloud 深度伪造检测Deepfake detection 声音水印Audio watermarking API平台API platform 企业向Enterprise-focused partial lifecycle: active 核验Checked 2026-10-30

Resemble AI 已从「声音克隆厂商」整体转向「多模态深伪检测与水印」企业平台(官网当日招牌在册): 检测(RAD-S1/DETECT-World)、水印(PerTh/C2PA)、身份注册三线;生成语音线转入开源研究归档、不再商用售卖。

Resemble AI has pivoted wholesale from “voice cloning vendor” to an enterprise multimodal deepfake-detection & watermarking platform (same-day headline on record): detection (RAD-S1/DETECT-World), watermarking (PerTh/C2PA), identity enrollment; the generative voice line moved to an open-source research archive and is no longer sold commercially.

适合与不适合When to use it, when not to

适合:企业反深伪(会议防护/客服欺诈/KYC/媒体水印)、AI agent 检测、C2PA 溯源合规。 不适合:找 TTS/声音克隆托管服务(商用语音线已撤)、个人免费使用、大陆直连备案入口。

Fits: enterprise anti-deepfake (meeting protection/support fraud/KYC/media watermarking), AI agent detection, C2PA provenance compliance. Not a fit: looking for hosted TTS/voice cloning (the commercial voice line was withdrawn), personal free use, direct mainland access with a filing.

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 国外主体(条款页当日在册文本为 "Resemble AI",全称后缀未在册;未见于网信办公示名单)。foreign entity (terms-page text same day reads "Resemble AI"; full legal suffix not on record; not in the CAC public registry).
国内使用Direct access
大陆可直连direct OK resemble.ai 与 /changelog、/terms-of-service、docs.resemble.ai 大陆直连实测 200(2026-10-30)。resemble.ai and /changelog, /terms-of-service, docs.resemble.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 Resemble 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 Resemble 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

自有检测模型线当日在册:RAD-S1(官方口径 first System One model for AI agent detection)、 DETECT-World(world model architecture for detection)、PerTh(多模态水印); 生成侧 Chatterbox 等语音模型转入 Research Archive(开源保留、不再作商用付费产品,docs 当日在册)。

Its own detection-model line on record same day: RAD-S1 (official wording: the first System One model for AI agent detection), DETECT-World (a world-model architecture for detection), PerTh (multimodal watermarking); on the generation side, Chatterbox and other speech models moved into the Research Archive (kept open-source, no longer offered as commercial paid products, on record in the docs same day).

运行位置Where it runs

云端 API 平台(docs.resemble.ai 当日在册:Detect/Watermark/Identity/Meetings 等端点)+ 即用面(Meetings 机器人入 Zoom/Teams/Google Meet、Chrome 扩展);本机不跑模型。

Cloud API platform (docs.resemble.ai on record same day: Detect/Watermark/Identity/Meetings endpoints) + ready-to-use surfaces (the Meetings bot joining Zoom/Teams/Google Meet, a Chrome extension); no local model.

本地文件Local files

未见官方声明的本机数据面;检测提交与结果在 Resemble 云端。Chatterbox 开源权重可本机跑(Research Archive 线)。

No locally-held data surface declared by the vendor was found; detection submissions and results live on Resemble's cloud. Chatterbox open weights can run locally (the Research Archive line).

关机后的任务Tasks after shutdown

API 任务形态(提交检测作业/轮询结果,docs 在册);Meetings 是实时入会检测形态。

API job form (submit a detection job / poll for results, on record in the docs); Meetings is a real-time in-meeting detection form.

工具与扩展Tools & extensions

产品面当日在册:Deepfake Detection(音视频图)、Watermarking(C2PA provenance)、Identity(声纹注册比对)、 Intelligence(可读结论)、Agent Detection、Audio Source Tracing;docs 提供 MCP server(docs.resemble.ai/_mcp/server 在册)。

Product surfaces on record same day: Deepfake Detection (audio/video/image), Watermarking (C2PA provenance), Identity (voiceprint enrollment & matching), Intelligence (human-readable verdicts), Agent Detection, Audio Source Tracing; the docs site offers an MCP server (docs.resemble.ai/_mcp/server on record).

上下文与记忆Context & memory

无会话概念——API 作业与注册身份(声纹/肖像)是持久面,作业间不共享上下文; 跨会话记忆不适用(检测平台形态)。

No session concept — API jobs and enrolled identities (voiceprint/likeness) are the persistent surfaces, and jobs share no context between them; cross-session memory does not apply (detection-platform form).

权限与限制Permissions & limits

企业向(Contact Sales 主打);身份注册面(Enroll and authenticate a voice)涉生物特征—— 授权与合规面(EU AI Act Article 50 合规清单资源页在册)是重点维度。

Enterprise-oriented (Contact Sales is the lead); the identity-enrollment surface (enroll and authenticate a voice) involves biometrics — consent and compliance (an EU AI Act Article 50 compliance-checklist resource page is on record) is the key dimension.

适合什么任务Best for

适合:企业反深伪(会议防护/客服欺诈/KYC/媒体水印)、AI agent 检测、C2PA 溯源合规。 不适合:找 TTS/声音克隆托管服务(商用语音线已撤,见 pitfalls)、个人免费使用、大陆直连备案入口。

Fits: enterprise anti-deepfake (meeting protection / support fraud / KYC / media watermarking), AI agent detection, C2PA provenance compliance. Not a fit: looking for hosted TTS/voice cloning (the commercial voice line was withdrawn, see pitfalls), personal free use, a filed direct-mainland entry.

头号误解Top misconception

  • 定位已整体转向——官网当日招牌是「Multimodal Deepfake Detection and Watermarking」,不是声音克隆厂商;docs 的 Voice Archive 页明言 Chatterbox 等生成模型「remain open and free…no longer offered as commercial paid products」——按旧印象把它当 TTS/克隆平台建档会整体错位The positioning pivoted wholesale — the same-day site headline is “Multimodal Deepfake Detection and Watermarking”, not a voice-cloning vendor; the docs Voice Archive page states Chatterbox and other generative models “remain open and free…no longer offered as commercial paid products” — profiling it as a TTS/cloning platform from old impressions mislabels it entirely
  • Voice Archive 文档保留但官方明言「do not imply continued availability of commercial voice services」——历史 API 文档在册≠服务仍在售Voice Archive docs are retained but the official text says they “do not imply continued availability of commercial voice services” — historical API docs on record ≠ the service still being sold
  • 条款页当日抓取未见 Inc. 主体后缀——主体全称未核验清,如实按在册文本写(未知就说未知)The terms-page fetch same day shows no “Inc.” entity suffix — the full legal name was not verified cleanly; recorded as fetched (say unknown when unknown)

未知项清单Known unknowns

  • 主体全称(条款页当日抓取未见 Inc. 后缀在册)
  • 商用语音线下架的具体时间与存量客户迁移安排(归档声明在册,细节未在册)
  • 检测 API 档位与价格(企业向 Contact Sales,未逐条核验)
  • 身份注册面(声纹生物特征)的数据留存与删除政策(privacy-policy 页未逐条核验)
  • The full legal entity name (the same-day terms-page fetch shows no "Inc." suffix on record)
  • The exact timing of the commercial voice line's withdrawal and migration arrangements for existing customers (the archive statement is on record; details not on record)
  • Detection API tiers and pricing (enterprise Contact Sales; not verified item by item)
  • Data retention and deletion policy for the identity-enrollment surface (voiceprint biometrics; the privacy-policy page not verified item by item)

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
officialResemble AI · 官网(检测/水印/身份产品线与 RAD-S1/DETECT-World/PerTh 模型线,2026-10-30 直抓 200)Resemble AI · official site (detection/watermarking/identity product lines and the RAD-S1/DETECT-World/PerTh model line, direct-fetched 200 on 2026-10-30)https://www.resemble.ai
officialResemble AI · Voice Archive(生成语音线转研究归档的官方声明,2026-10-30 直抓 200)Resemble AI · Voice Archive (official statement moving the generative voice line to an open-source research archive, direct-fetched 200 on 2026-10-30)https://docs.resemble.ai/voice-archive
officialResemble AI · Terms of Service(2026-10-30 直抓 200)Resemble AI · Terms of Service (direct-fetched 200 on 2026-10-30)https://www.resemble.ai/terms-of-service
changelogResemble AI · Product Changelog(官方更新日志,2026-10-30 直抓 200)Resemble AI · Product Changelog (official change log, direct-fetched 200 on 2026-10-30)https://www.resemble.ai/changelog
docsResemble AI · docs(API 平台文档含 MCP server 入口,2026-10-30 直抓 200)Resemble AI · docs (API platform docs including the MCP server entry, direct-fetched 200 on 2026-10-30)https://docs.resemble.ai/welcome

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):

  • 深伪检测的假阳性率(官方有 Benchmarks 页自称可比,第三方未复核——检测类产品的生死线指标)
  • Meetings 机器人入会的实际延迟与覆盖(Zoom/Teams/Google Meet 三平台在册,实测面未核验)
  • Chatterbox 开源权重与已撤商用托管线的功能差(归档声明在册,差异未逐条核验)

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):

  • False-positive rate of deepfake detection (the official Benchmarks page self-claims comparisons; no third-party re-check — the make-or-break metric for a detection product)
  • Actual latency and coverage of the Meetings in-meeting bot (Zoom/Teams/Google Meet on record; the measured surface not verified)
  • Feature gap between Chatterbox open weights and the withdrawn hosted voice line (the archive statement is on record; the gap not verified item by item)

相关条目Related entries

  • ElevenLabsElevenLabs镜像对照:ElevenLabs 卖语音生成+克隆并附带 AI 检测面,Resemble 反过来把检测做成主业、生成线退归档。A mirror-image contrast: ElevenLabs sells speech generation + cloning with an AI-detection side feature; Resemble flipped it — detection is the main business, the generation line went to the archive.
  • AssemblyAIAssemblyAI同为音频 API 平台:AssemblyAI 做转写理解,Resemble 做真伪鉴别。Also an audio API platform: AssemblyAI does transcription and understanding, Resemble does authenticity adjudication.
  • HeyGenHeyGen检测面覆盖合成视频:HeyGen 这类数字人视频正是深伪检测要鉴别的对象形态之一。The detection surface covers synthetic video: HeyGen-style digital-human video is exactly one of the object forms deepfake detection has to adjudicate.

本页由This page is generated from the ai-handbook 数据层生成(CC BY 4.0)。data layer (CC BY 4.0). 方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.