-changed python requirement from 3.7+ to 3.8+ (following Whisper)
-more reliable word-level timestamps (using Whisper's new method for word timestamps)
-`transcribe()` now returns `WhisperResult` object (allowing easier to manipulation of results)
-`WhisperResult` contains methods to save result as JSON/SRT/VTT/ASS
-`WhisperResult` contains methods to regroup segments word by word
-added [Silero VAD](https://github.com/snakers4/silero-vad) for generating suppression mask (requires PyTorch 1.2.0+)
-improved non-vad suppression
-added `visualize_suppression()` for visualizing suppression based on arguments (requires Pillow or opencv-python)
-SRT/VTT/ASS outputs now all support both segment-level and word-level