Add support for speech recognition via the OpenAI-compatible API

We would like to add support for OpenAI-compatible speech recognition APIs and the ability to deploy our own speech recognition models.

More importantly, add more vendors, such as Zhipu