Draft: Basic example server

[Azeirah]

A server example for whisper.cpp, based on the LLaMa.cpp server example.
This is still a work-in-progress, but the very basics work.

Why would you want to run whisper as a server? If you need to process a short clip often, for example on a voice-controlled application, the model you're using is loaded into RAM every time ./main is invoked. For very short voice snippets of about 3-5 seconds, about 50% of the time is spent loading the base-en model every time.

With this you can simply send a http request to this web-server and your wav file gets converted immediately.

Progress

• Supports basic (model selection & performance & quality) parameters for whisper
• Supports same server parameters as LLaMa.cpp server example
• Supports /convert endpoint, send a .wav filename and get transcription + metadata back as JSON
• Sends error messages in JSON when something goes wrong
• Supports diarization
• Supports language detection
• Server is documented in README.md

Note that it's been a while since I used C++, I made most of this happen with frankenstein copy-pasting. If I'm doing something stupid please let me know.

[litongjava]

When do you plan to merge this PR, I want to use this feature as soon as possible.

[Azeirah]

When do you plan to merge this PR, I want to use this feature as soon as possible.

Whenever ggerganov is done with his changes. I'll make sure a reasonably feature complete server example will be merged.

#1369 (comment)