Speechmatics offers speech APIs for developers who need reliable transcription in messy, real-world audio. It is designed to cope with different accents, several people talking and multilingual speech, in real time or from recordings.
Along with speech-to-text, it provides text-to-speech and a model aimed at voice agents, plus documentation and integrations. You can test it with samples or your own voice in the browser demo, and get started for free before moving to paid usage.