Verbose Response should include exact (not integer rounded) timestamps

[samhincks]

Hello there, thanks for this API!

The one problem I have is that the verbose response gives integer rounded timestamps for the positions of different speech. For my use case, I need the exact moment a fragment of speech happens... The error can be seen below. "Let's see..." doesnt start at exactly 20.0 or end at 32.0..

Would be very grateful if this was fixed :)

{
      "avg_logprob": -0.28310803266671986,
      "compression_ratio": 0.9714285714285714,
      "end": 32.0,
      "id": 7,
      "no_speech_prob": 0.004850671626627445,
      "seek": 2000,
      "start": 20.0,
      "temperature": 0.0,
      "text": " Let's see... Let's see about that.",
      "tokens": [
        50364,
        961,
        311,
        536,
        485,
        961,
        311,
        536,
        466,
        300,
        13,
        50964
      ],
      "transient": false
    }
  ],