I’ve been working on building a chatbot, and one of the features I want to include is speech-to-text. Since I’m part of a startup, budget is definitely a constraint. At the same time, due to security and compliance requirements, I’d prefer to avoid relying on external APIs.

For an MVP or pilot launch, I’m trying to figure out which ASR approach or architecture would make the most sense to start with. I’ve been looking into options like Whisper, Parakeet, etc., but I’m a bit unsure about the best starting point given my constraints but also having the low latency criteria.

Would really appreciate any suggestions or insights from people who’ve worked on something similar, especially around trade-offs between self-hosted models vs APIs, performance, and ease of deployment (I am ready to take on the challenge for deployment).