I built the site Mati which shows one sentence a day; the goal is to repeat the sentence and get feedback. TLDR Cloud Text-to-Speech offers natural voices, way better than the boring robotic voices browsers have by default. Google offers a free tier which is enough for experimenting with the service. Speech-to-Text while not new, now most browsers add an LLM layer which tries to understand what was said, making for better transcripts. However, if you are looking for feedback on your pronunciation, you can use a speech model ( wav2vec2 or whisper ) only for the transcription part, without the guessing layer. Back End There's a daily cron that runs two functions to generate a sentence and get its corresponding audio. Then the site is deployed using the new sentence.…