how do you decide what to talk about when you walk

the language training app - something’s wrong with it and it’s not just the need for text instructions it’s something more fundamental.

2 45s vocal language classes a day, play anytime you want mini games, sundays you get feedback from a trainer. curriculum constantly adapts to your level and needs.

Would be helpful if the app showed the text as I spoke to it and helped me refine how to correctly pronounce things or give me tips on how to improve or retain vocabulary

Stay Curious // iixii // Head of Crypto at https://www.lazertechnologies.com/  // Prev: CoinDesk, Roc Nation // based @basedevo

so the issue is rn it’s powered by openai’s realtime api which abstracts the ai part to a webrtc call, however that doesn’t include native transcribing, if you request transcriptions it uses whisper which is often wrong - i can either pray meta generates a voice model, or i can setup a pipeline of audio in -> transcribe -> llm -> speech gen, which sounds not fun unless i can find an OSS repo to automate it