一个输入法卖 1000 块,我做了个免费版An input method sells for 1000 yuan, I made a free version


AI free token/AI免费积分

图片

我以前特别不喜欢语音输入。

微信的语音转文字、搜狗、苹果自带的听写,都试过。体验基本一致:错别字一堆,标点乱飞,说话中间停顿一下思考,出来的文字就没法看。更别说讲话本身就是跳跃的,想到哪说到哪,转写出来的东西自己都不想读第二遍。

所以我一直靠打字。日记、聊天、给 AI 下指令,全靠键盘,打字速度也练出来了。

今年年初,朋友们开始分享一款叫 Typeless 的 AI 语音输入法。我下载试了一下。

完全不一样。不管中间怎么停顿、说错词、语序混乱,它都能整理成你本来想说的话。我只管讲,出来的文字比我自己打的还清楚。

然后我看了一下价格。月付 200左右,年付 1000多。

一个输入法,1000 块。而且真有不少人在买。

那段时间我刚好在琢磨用 AI 做产品,看到这个就想:这东西我能不能自己做一个?

于是走上了这条不归路。

研究实现逻辑,测了几十个语音识别模型组合,提示词改了不知道多少版。中间无数次觉得"差不多了",换个场景又不行。反反复复,前前后后折腾了好几个月。

最终做出来了,叫 Typefree。内测的朋友用完反馈,有些场景效果甚至比 Typeless 还好。

我自己也实打实地用了下来,到现在累计输入近 20 万字。输出质量上,基本平替了 Typeless。

成本呢?豆包和阿里API 平台送的免费额度,我高强度用了三个多月还没花完一半。就算额度用光了自己充值,高强度用一年也就几十块。

1000 块变成几十块,这个账怎么算都划算,而且数据全在自己本地更放心。

说这么多,到底用起来什么感觉?直接看几个场景。

图片

我习惯睡前对着电脑随便说两句当日记,想到什么说什么。以前语音转写出来就是一团乱麻。现在它会自动去掉废话和语气词,保留原话原意,排版干净。

图片

说错了也没关系。有一次我把名字说错了,中间又纠正了一下,它最终输出的就是正确的版本。不会把错的和对的都留着让你自己挑。

图片

给 AI 下指令的时候感受更明显。脑子里的逻辑是清楚的,说出来经常东一句西一句。它会帮你把意思梳理成结构化的表达,直接就能丢给 AI 用。指令越精准 AI 干活越好,这一步省了不少来回。

图片

数字也会自动处理,该用中文还是阿拉伯数字,它根据场景自己判断。

图片

另外一个挺实用的是热词词库。常用的人名、产品名、专有名词,加进去说话时就能准确识别。默认内置了一批科技词汇,像 Claude Code、Anthropic、Openclaw、ChatGPT,或者姚晓媛,这句话也是用软件一次输出的,它都能精准识别。

图片

关于隐私:数据全在你本地,不过我的服务器。没有用户登录系统,不存任何东西。自己配了 API Key 之后,请求直接发给服务商。

版本很简单。永久免费版每天 1500 字,自己配个 API 就能用,5 分钟搞定。新用户还有 7 天免配置试用,下载直接体验。觉得不够用可以一次性买断,没有日输入上限。,没有订阅制。

感谢朋友们和内测用户的支持和建议💗

强烈建议下载体验一下

免费下载地址:typefree.app

图片

 

图片

I used to hate voice input.

I tried WeChat's voice-to-text, Sogou, and Apple's built-in dictation — they're all terrible. Typos everywhere, punctuation is a mess, and if you pause to think while speaking, the output is unreadable. Let alone when your speech is naturally fragmented — the resulting text is barely worth reading.

So I've always relied on typing. Diary entries, chats, AI prompts — all done via keyboard. I got pretty fast at it.

Early this year, friends started sharing Typeless, an AI voice input app. I gave it a shot.

It was completely different. No matter how I paused, misspoke, or scrambled the order, it organized the text into what I actually meant. I just had to speak, and the output was clearer than what I could type myself.

Then I saw the price: ~¥200/month, ~¥1000/year.

A thousand yuan for an input method? And people are actually buying it?

At that time, I was working on AI products. I thought: could I build this myself?

That's how I started down this path.

I researched the underlying logic, tested dozens of speech-to-text model combinations, and wrote countless versions of prompt parameters. Many times I thought "this is good enough," only to find it failing in new scenarios. This went on for months.

Finally, I succeeded. It's called Typefree. Testers said it works as well as — or better than — Typeless in many cases.

I've used it extensively for nearly 200,000 words. Quality-wise, it matches Typeless.

Cost? Doubao and Alibaba API platforms gave me free quotas that lasted over three months of heavy use. Even if I run out and pay, a year's worth of heavy usage would only cost tens of yuan.

From a thousand yuan to tens of yuan — that's a no-brainer. Plus, all data stays local — safer and more private.

What does it actually feel like to use? Let me show you some scenes.

图片

Before bed, I like to ramble into the computer as a diary — whatever comes to mind. Previously, the transcription was a mess. Now, it automatically filters out filler words and reorganizes the content logically, maintaining my original intent with clean formatting.

图片

If you make a mistake, no problem. Once I mispronounced someone's name and then corrected myself mid-sentence — the final output captured only the correct version, not both.

图片

When giving AI instructions, this is especially useful. Your brain knows what it means, but sometimes your mouth doesn't. The tool helps structure the text clearly, so you can send accurate, executable prompts to the AI without rewriting.

图片

It also handles numbers automatically — switching between Chinese characters and Arabic numerals based on context.

图片

The hotword library is great. You can add common names, product codes, or technical terms so they're recognized accurately during speech. For example, "Claude Code," "Anthropic," "Openclaw," "ChatGPT," or my name "Yao Xiaoyuan" — all are correctly identified.

图片

About privacy: all data is processed locally. There's no user login system, and nothing is stored. You configure your own API key, and requests go straight to the service provider.

There are two versions. The free version limits you to 1500 words per day — easy to set up with your own API key. For a low one-time purchase, you get unlimited word count with no subscription fees. There's also a 7-day free trial if you want to test before committing.

Thanks to friends and beta users for the feedback!

Seriously, try it — download here: typefree.app

图片

图片

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry