/PhyzAI

Primary LanguagePython

Notes (windows 10)

  • check device index
import speech_recognition as sr
for index, name in enumerate(sr.Microphone.list_microphone_names()):
    print("Microphone with name \"{1}\" found for `Microphone(device_index={0})`".format(index, name))
  • torch cuda is default to fp32 if fp16 is not specified
  • Some faster inference model than openai/whisper:
    • faster-whisper fp16 is said to be 4x faster than openai/whisper
      • tested on my computer (i5-12600K, RTX 3060, NVidia v522.04, CUDA 11.8)
        • 0.88 - 1.02 seconds
    • whisper-cpp