Thoughts on the development of SpeakAI

The initial idea can be traced back to the end of February and the beginning of March. At that time, ChatGPT became popular in China, and there were many websites with direct shells on the market. At that time, I thought that ChatGPT+ voice technology could be well implemented in oral language learning. .

Unexpected surprise

I learned Taro a bit, then spent 2 weekends developing it, and launched the first version of the mini program at the end of March. I posted a post on Xiaohongshu, but I didn’t expect the market feedback to be good, far exceeding expectations and accumulating the first wave of users.

First lesson learned

But I didn’t expect it to be reported and taken off the shelves after 3 days of going online, and I felt very angry at the time. This is the first lesson learned. The competition you think is a security attack, but the real competition is a malicious report. Later, there was a wave of complaints with the review of WeChat. First, the dialogue function was temporarily removed, and only a pronunciation evaluation function was added. It took a while to add the AI dialogue feature back.

Later, when the app was launched, some people posted bad reviews maliciously in the app store. At this time, I just mocked and went back, and then the other party deleted the bad reviews, which is very funny.

APP Development

Because of the small program problem, the development progress of APP was accelerated. I bought a mac mini, learned a bit about React Native, tossed about the APP development environment, and then launched the IOS and Android versions at the end of April.

The feedback after the APP was launched is also very good. According to user feedback, version 1.0.1 focused on optimizing a wave of dialogue experience, such as translation/prompt words/notebooks. In version 1.0.2, the real-time mode is launched, which can realize the effect of making a phone call. It was already mid-May at this time.

Operation Promotion

I don’t know this at all, but I learned a lot by watching the promotion methods of similar products.

For example, the activity of “Sending Xiaohongshu to members/App Store praise to members”, as well as the promotion method of agency/secondary distribution, I didn’t expect to be able to play like this, and I gained a lot of posture. It turns out that many kinds of cursive/good reviews come from this way~

However, because SpeakAI does not charge, none of these methods can be used. You can give it a try when a paid plan is launched later.

Next step planning

After 3 versions of iterations, the dialogue experience is basically OK. Because I only have to do it on weekends, and I don’t have much time. I thought that competing products would go deeper in functions/scenarios, but I didn’t expect that competing products would spend all their attention on operation and promotion, and the functions would not be much after the first version. Iterated, but SpeakAI has done more in terms of functions.

At this time, I handed over a lot of development work to my friend, and I can devote my energy to research and thinking.

In the next step, we should still focus on oral English learning scenarios and provide a more complete learning experience, because AI dialogue is only a capability, and it will eventually fall into learning to really help users learn oral English better. Personally, I feel that relying on nonsense dialogue alone is of limited help to oral English learning. Early users may find it fresh, but if it is of no value, it will become cold after a while. This is also a point worthy of further thinking and development, and it is also a place that is really difficult and has barriers.

SpeakAI开发的感想

最开始的想法要追溯到2月底3月初的样子，那时候ChatGPT在国内火起来不久，市面上比较多的还是直接套壳的网站，当时想着ChatGPT+语音技术可以很好地在口语学习上落地。

意外的惊喜

学了一下Taro，然后花了2个周末开发，在3月底上线了第一版小程序。在小红书上发了个贴，没想到市场反馈还不错，远远超出预期，积累了第一波用户。

学到的第一课

但是没想到上线3天就被人举报下架了，当时感觉很气。这是学到的第一门课，你以为的竞争是安全攻击，真实的竞争是恶意举报。后面跟微信的审核申诉了一波，先暂时把对话功能去掉，只上了一个发音评估的功能。过了一段时间才把AI对话的功能加回来。

后面上线APP，在app store也有人恶意刷差评，这时候我直接嘲讽了回去，然后对方把差评删了，很搞笑。

APP开发

因为小程序的问题，加速了APP的开发进度。买了台mac mini，学了一下React Native，折腾APP开发环境，然后在4月底上线了IOS和Android版本。

APP上线后反馈也很好，根据用户的反馈，1.0.1版本集中优化了一波对话体验，比如翻译/提示词/笔记本。而在1.0.2版本，上线了实时模式，可以实现打电话的效果。这时候已经5月中旬。

运营推广

这块我是完全不会，看同类产品的推广方式倒是学到很多东西。

比如“发小红书送会员/App Store好评送会员”的活动，还有代理/二级分销的推广方式，没想到还能这样玩，涨了不少姿势。原来很多种草文/好评都是这样来的～

不过因为SpeakAI没收费，这些方式都用不了。后面推出付费计划了倒是可以试一试。

下一步规划

经过3个版本的迭代，对话体验上基本是OK了的。因为我只有周末弄一弄，没太多时间，本来以为竞品在功能/场景上会做的更深，但没想到竞品的心思都花在运营推广上，功能在第一版后就没怎么迭代了，反而SpeakAI在功能上做得更多。

这时候我把很多开发工作都交给了我朋友，而我可以把精力投入到调研和思考上。

下一步，应该还是会专注口语学习场景，提供更加完整的学习体验，因为AI对话只是一个能力，最终还是要落到学习上，真正帮助用户更好地学习口语。个人感觉光靠瞎对话，对口语学习的帮助还是有限，早期用户可能觉得新鲜，但是没有价值的话，过一段时间就会凉。这也是值得去进一步思考和开拓的点，也是真正困难且有壁垒的地方。