Whisper API not compatible with OpenAI

[mutschler]

LocalAI version:
localai: stable 2.28.0 (bottled), HEAD (via brew)

Environment, CPU architecture, OS, and Version:
Darwin MacBook-Pro.local 23.3.0 Darwin Kernel Version 23.3.0: Wed Dec 20 21:30:59 PST 2023; root:xnu-10002.81.5~7/RELEASE_ARM64_T6030 arm64

Describe the bug
The whisper api endpoint (/v1/audio/transcriptions) isn't compatible to OpenAI's implementation.

While support for response_format (and other request parameters) would be amazing i think at least the default response should be compatible with OpenAI's verbose_json.

To Reproduce
spin up localai, install whisper-1 model, send any audio file to /v1/audio/transcriptions and check the response

Expected behavior
Get a response compatible with https://platform.openai.com/docs/api-reference/audio/verbose-json-object

Additional context
i know that OpenAi defaults to json for response_format but since the parameter isn't supported and the current answer is already close to the verbose_json format changing it so one could use official libraries or use localai as drop in replacement

The main differences i spotted:

task, language and duration are missing from the top level
seek, temperature, avg_logprob, compression_ratio and no_speech_prob are missing from each segment and the start and stop values seem to be (rounded?) Nanoseconds instead of Seconds (5000000000 instead of 5.119999999999998 for around 5 seconds)