Add fields to verbose_json response and show examples on the home page

[JacobLinCool]

The verbose_json format seems to aim to match the same format as OpenAI's, but it is currently missing several fields.

• I tried to include task, language, duration, temperature, and avg_logprob.

Some observations:

• The current segment.words do not match OpenAI's (OpenAI does not provide this one), but I think that's okay; it can be an extension to showcase the capabilities of whisper.cpp at the word level.
• It seems that OpenAI also includes non-speech tokens.
• For compression_ratio and no_speech_prob, if I have not missed anything, currently, they cannot be obtained from whisper.cpp?

OpenAI's verbose_json

{
  "task": "transcribe",
  "language": "english",
  "duration": 4.440000057220459,
  "text": "This is my voice sample for research purpose only.",
  "segments": [
    {
      "id": 0,
      "seek": 0,
      "start": 0.0,
      "end": 4.0,
      "text": " This is my voice sample for research purpose only.",
      "tokens": [
        50364,
        639,
        307,
        452,
        3177,
        6889,
        337,
        2132,
        4334,
        787,
        13,
        50564
      ],
      "temperature": 0.0,
      "avg_logprob": -0.4242081940174103,
      "compression_ratio": 0.8928571343421936,
      "no_speech_prob": 0.0008698556339368224
    }
  ]
}

whisper.cpp's verbose_json

{
  "task": "transcribe",
  "language": "english",
  "duration": 4.440000057220459,
  "text": " This is my voice sample for research purpose only.\n",
  "segments": [
    {
      "id": 0,
      "text": " This is my voice sample for research purpose only.",
      "start": 0.0,
      "end": 4.0,
      "tokens": [
        770,
        318,
        616,
        3809,
        6291,
        329,
        2267,
        4007,
        691,
        13
      ],
      "words": [
        {
          "word": " This",
          "start": 0.16,
          "end": 0.42,
          "probability": 0.8665725588798523
        },
        {
          "word": " is",
          "start": 0.42,
          "end": 0.63,
          "probability": 0.9934033751487732
        },
        ...,
        {
          "word": ".",
          "start": 4.0,
          "end": 4.0,
          "probability": 0.8532010316848755
        }
      ],
      "temperature": 0.0,
      "avg_logprob": -0.1288750171661377
    }
  ]
}

I also replaced the hello on the homepage with request examples. I think the first action after running the server with ./server is to open the URL in the terminal and check out the homepage. It may be easier for users to try the server this way. (A web interface for direct interaction would be better, but it requires some time to design...)

[bobqianic]

For compression_ratio and no_speech_prob, if I have not missed anything, currently, they cannot be obtained from whisper.cpp?

Indeed, we opt for entropy rather than gzip compression. Currently, I'm working on implementations related to the no_speech_prob

[JacobLinCool]

Currently, I'm working on implementations related to the no_speech_prob.

Wow! That sounds great! I believe it can definitely improve the use of whisper.cpp.

I think we can merge this pull request first and then either add a new issue for that or include some TODO comments in the code. What do you think?

[JacobLinCool]

I've added a form to the home page.

[JacobLinCool]

The problem may not related to the changes in this PR, which only changed the frontend page and the output format. The mechanism of file processing is not involved.

I think the problem is about the changes in #1781, which allows passing audio file content as file name. (I think that's why you got error: failed to open 'RIFF$�' as WAV file)

[UniversalTechno]

sorry it was compilation mistake from my side thanks