实时通话
全双工，默认 grok-voice-latest。嘈杂时可以改成按住说话。
免费 AI 语音 · 同一页
Free AIvoice Online — 免费 AI 语音工作台
把文稿读成配音，或把录音收成可以编辑的文字。NorSpeak 把朗读、转写和实时通话放在浏览器里，不用安装软件。
无需安装 · 在线免费试用 · MP3 朗读 · 中文、English、日本語
Your words 你的话
Make every word count. 让每个字都很重要。
12 / 8000
可以粘贴上游 Key，或切换到已配置的 Worker 密钥。
字和声音，留在同一页。
在线工作室走部署上的密钥并有公平使用限制。外部 API 使用账户里的 NorSpeak Key，并扣减额度。粘贴的上游 Key 仍会原样转发。
Why free aivoice here
NorSpeak is a free aivoice workbench you can use in the browser. Get free AI voice text-to-speech, live duplex calls, captions, and speech-to-text on one page—no app install. Paste text to hear free AI voice samples, or open the studio for realtime conversation. Built for creators who want free aivoice tools that stay fast, clear, and ready for API use when you need more quota.
给创作者、客服和开发者使用。实时通话、转写和朗读，都留在这一页。
NorSpeak keeps listening, speaking, and transcribing on a single page, so you can judge the sound before you commit to an integration. Open the site, paste a sentence, and hear it through the same read-aloud route the studio uses. If the line feels flat, switch the voice or the language and play it again. The browser tab is the product. There is no desktop installer and no separate editor to learn.
A live call is the next step on this same address. You speak, captions land beside the audio, and the transcript stays in the tab until you export text, subtitles, or JSON. Creators use that path for a product walkthrough, a short interview, or a lesson voiceover. Support teams use it to hear how a reply will sound before a customer does.
Speech-to-text sits next to playback. Upload a recording or dictate in the workbench, then send the words back through the reader when you want a second pass. Chinese, English, and Japanese share one key. When a demo needs more quota, the same three routes—read aloud, transcribe, and realtime—are the ones your own worker calls. This page is where you learn their shape. The key stays in a worker secret or in this browser, and the public site only forwards the request.
Free AIvoice 能覆盖这些场景
查看能力- 实时通话
- 字幕记录
- 文件转写
- 摘要朗读
- 对话导出
- 六种音色
- 中文
- English
- 日本語
- 全双工
- 按住说话
- MP3
- SRT
- TXT
- JSON
- 语音接口
- 工作台
- 同一把 Key
两套用法建在同一条语音链上
NorSpeak 工作室
把通话、字幕和导出留在同一页。
NorSpeak 听写
把字读成声音，也把声音收成字。
NorSpeak 工作室首页
今天想留下什么？
通话字幕朗读导出
发布说明
访谈记录
课程旁白
NorSpeak 听写示意
这场对话还在同一页
你 · 把今天的发布收成三句。
NorSpeak · 实时在左边，字幕在右边。
NorSpeak 工作室
在同一页完成 free AI voice 通话、转写和朗读
开口留下实时通话，旁边写出字幕，再把摘要读回去。中文、英文和日文，用同一把 NorSpeak API Key。
把今天的发布收成三句。实时通话留在左边，字幕留在右边，摘要可以再读一遍。
通话和字幕
实时开口，记录留在同一页。
把今天的发布收成三句
播放请求本站 POST /api/tts，由 Worker 转发给 NorSpeak。
字幕导出
TXT、SRT、JSON。摘要在浏览器里抽出，再交给朗读。
六种音色
ara、eve、leo、rex、sal、luna
中英日
language 传 zh、en、ja，或 auto。
一次会话怎么在这一页走完
A useful session stays on this address. You do not create an account here, and you do not download a recorder. The first check is whether a sentence sounds right. The orb in the capability panel plays a fixed Chinese line through the read-aloud route. The reader card lower on the page takes your own text, a language, and one of six voices, then returns an MP3 the same way.
When the line is right, open the studio. The hash changes to studio and the document stays on the homepage. Allow the microphone. The default call is full duplex on the flagship voice model. If the room is noisy, switch to hold-to-talk so the microphone is only open while you press. Captions land beside the call. A file transcription uses the same record as the live captions, so you are not copying between tools.
Export stays in the browser: plain text, SRT subtitles, and JSON. A short summary can be taken from the transcript and sent back through the reader if you want to hear the recap. The audio request, the file upload, and the live socket all go to this site first. The worker forwards them. The browser never puts the server key on the upstream URL. A pasted key stays in session storage unless you choose to remember it on this computer. Leave that box unchecked on a public machine. An optional access token only unlocks a worker-held key for the current tab. It is not the voice key.
Six built-in voices cover the reader. Ara is warm and friendly, Eve is a bright British voice, Leo is steadier, Rex is clear and confident, Sal is smooth and even, and Luna is patient and soft. Language is separate: Chinese, English, Japanese, or automatic. Automatic helps when a paragraph mixes languages. A lesson that must stay in one language should set the code so the reading does not drift.
The model stack lists the realtime names. The latest alias always points at the current flagship. The older 2.0 name is kept so existing clients still resolve. File transcription has its own default model. The read-aloud route does not take a model name. It needs text and a language, and a single request stops at 8000 characters, so split a long script. Try the call in the studio, then repeat that shape from your own worker. The page and the API are one chain.
这一页把试听、通话和转写放在一起，适合先确认声音，再决定要不要接到自己的程序里。
先听。能力区中间的圆钮会播放一句固定的中文。再往下，朗读卡片可以贴上你自己的文字，选择中文、英文、日文或自动，并换成六种音色之一。单次朗读有长度上限，更长的稿子分成几段即可。
再通话。打开工作台后地址仍然是首页。允许麦克风之后，默认是全双工。环境吵的时候可以改成按住说话。字幕和文件转写写在同一条记录里，结束后可以导出纯文本、字幕或 JSON。摘要在浏览器里抽出，再交给朗读听一遍。
最后看接口和密钥。浏览器只请求本站，再由转发层送到上游。服务器上的密钥不要写进仓库。粘贴到页面的密钥默认只留在这次会话，公共电脑不要勾选记住。访问令牌只用来放开服务器上的额度，并不是语音接口的密钥。
创作者、客服和做集成的人可以按这个顺序使用：浏览器里就能把 free aivoice 走完，先听一句，再留下一场对话，确认无误后再接 API。
Who uses this page
Three kinds of work fit on this page, and they can happen in one sitting.
A creator usually starts with a sentence that has to sound like the product. Paste the line into the reader, pick a language, and step through the voices until the tone is right. Ara stays warm, Rex stays direct, and the others sit between them. The playback you hear is the read-aloud route, so a later API call does not surprise you with a different engine.
A support lead or interviewer needs the words, not only the sound. The studio keeps the call and the captions in one record. Full duplex is the default. Hold-to-talk is the fallback when a room is loud. Export text, subtitles, or JSON when the conversation is finished, and read a short summary back if you want to check the recap by ear.
Someone wiring the API can watch those same steps and then read the resource section. The read-aloud call is JSON. Transcription is a file upload. The live socket begins with a hello frame and does not put the key in the address. Once the studio behaves the way the integration should, the worker in front of this page is the client to copy.
外部调用使用付费 API Key
定价与 Key文本转语音
页面上的试听走免费在线额度。下面的示例是带 NorSpeak Key 的付费调用。
语音转文字
POST /api/stt，默认 grok-voice-transcribe-2.0。文件放在表单里，由 Worker 原样转发。
实时通话
WebSocket /api/realtime。第一帧是 workbench.hello，Key 不写进上游地址。
await fetch("https://norspeak.com/api/tts", {
method: "POST",
headers: {
authorization: "Bearer nsk_...",
"content-type": "application/json",
},
body: JSON.stringify({
text: "让语音技术活起来。",
language: "zh",
voice_id: "ara",
}),
});现在接上的语音栈
01 · 实时
grok-voice-latest
始终指向当前旗舰
02 · 实时
grok-voice-3.0
旗舰语音模型
03 · 实时
grok-voice-2.0
兼容旧模型名称
04 · 转写
grok-voice-transcribe-2.0
文件上传和独立听写的默认模型。
05 · 朗读
六种内置音色
ara 温暖、友好 · eve 轻快英音 · leo 沉稳英音 · rex 清晰、自信 · sal 平滑、均衡 · luna 耐心、柔和
免费在线，付费接口
NorSpeak API Key 只存哈希。明文只在创建时出现一次。
没有 Key 的浏览器请求必须同源，并受公平使用限制。
付费 Key 会检查套餐额度。TTS 按字符，STT 按转写字符，实时按分钟。
常见问题
What is free aivoice?
It is the browser workbench on this site: text-to-speech, speech-to-text, and a live duplex call with captions, kept on one page. There is no desktop app to install. Paste a line to hear a sample, or open the studio when you want the conversation written down beside the audio. Chinese, English, and Japanese can share that session. When you need more quota, the same routes are available from your own worker.
Is NorSpeak free AI voice?
Yes. Open this page and try the workbench in the browser: paste a line, play it, and start a live call with captions. A NorSpeak API key authorizes the voices. Keep that key in this browser, or in a worker secret if you deploy the page yourself. It does not belong in the repository. When a project needs more quota, call the same read-aloud, transcribe, and realtime routes.
How to start free aivoice online?
Start on this page. Play the sample in the capability panel, or scroll to the reader and paste your own line. Choose a language and a voice, then listen. To talk instead of type, open the studio, allow the microphone, and speak. Captions stay next to the call. Export the transcript when you are done, or read a short summary back. The resource section shows the same three routes if you later call them from your own app.