One thing that really frustrates me about using LingQ is the quality of the transcriptions when you import audio. It’s not just because the speaker talks fast or there’s background noise, the problem is that the transcription is often flat-out wrong in terms of phonetics. You get characters that sound similar but make no sense in context, and it throws off your understanding completely.
So what I’ve started doing is printing the lesson text before starting, and sending it to Deepseek. I tell it clearly that the text comes from a transcription and probably has phonetic errors, and I ask it to correct them using context. It does a much better job than ChatGPT or Gemini for this kind of thing.
What’s even better is that Deepseek usually tells you what it changed and why, and that’s when you realize just how many mistakes there are in LingQ’s transcriptions. After that, all I do is copy the corrected text, go back to LingQ, hit “Edit Lesson”, then click “Regenerate Lesson” at the bottom left, delete the existing content, and paste in the new one.
It’s a bit of extra work at first, but totally worth it.
5 Likes
This is because Deepseek was developed by Chinese research team.
2 Likes
Yeah, I know the app is Chinese. It’s just a shame it doesn’t have a microphone interface, it would make things a lot smoother by skipping the need to type out every request. But still, if you’re focused on Chinese, it works really well.
To be honest, I’m a bit fed up with typical Chinese learning videos. After watching hundreds, they all start to feel kind of the same. So I started looking for something more engaging that could also serve as solid comprehensible input.
Now what I do is search for random English videos on topics I actually enjoy. I grab the transcript and throw it into ChatGPT or Gemini. I tell it I’m a Spanish speaker and ask it to rewrite the content in simplified Spanish, something around A2 or B1 level. Since it’s my native language, I can read it easily and fine-tune the tone if needed, whether I want it more casual, less literary, whatever.
Once I’m happy with it, I ask for a Chinese translation in the same tone. And here’s where it gets interesting. I paste the Chinese version into Deepseek. I let it know the text was generated by another AI, not native in Chinese, and I ask it to rewrite it so it sounds natural to a native speaker. Nothing literal, just clean and native-sounding.
Once that’s done, sometimes I add a TTS voice. With the refined text and the audio, I create a mini lesson for myself. It would honestly be much easier if LingQ had built-in TTS, but for now, this setup works great.
2 Likes