{\rtf1\ansi\ansicpg1252\cocoartf2639
\cocoatextscaling0\cocoaplatform0{\fonttbl\f0\fswiss\fcharset0 Helvetica;}
{\colortbl;\red255\green255\blue255;}
{\*\expandedcolortbl;;}
\margl1440\margr1440\vieww11520\viewh8400\viewkind0
\pard\tx720\tx1440\tx2160\tx2880\tx3600\tx4320\tx5040\tx5760\tx6480\tx7200\tx7920\tx8640\pardirnatural\partightenfactor0

\f0\fs24 \cf0 ######################################################################\
#  Instantiate a model\
# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\
#\
# load OpenAI Whisper automatic speech transcription \
import whisper\
import subprocess\
import librosa\
\
\
import argparse\
from whisper.utils import str2bool\
parser = argparse.ArgumentParser()\
parser.add_argument('--audio_file', default=\'91./test.wav')\
parser.add_argument('--modelsize',default='small.en')\
#parser.add_argument('--fp16',type=str2bool,default=1)\
parser.add_argument("--fp16", type=str2bool, default=True, help="whether to perform inference in fp16; True by default")\
args = parser.parse_args()\
\
print('===========================================')\
\
audio_file = args.audio_file\
modelsize = args.modelsize\
fp16flag = args.fp16\
\
\
audio_len = librosa.get_duration(filename=audio_file)\
print('Audio Length(sec):',audio_len)\
\
waveform, sr = librosa.load(audio_file)\
print('Audio sampling rate(hz):',sr)\
\
\
if fp16flag:\
  print('FP16 Half-precision Model Loading')\
else:\
  print('FP32 Full-precision Model Loading')  \
\
def ML_model_load():\
\
  # choose among "tiny", "base", "small", "medium", "large"\
  # see https://github.com/openai/whisper/\
  print('Loading whisper model:')\
  print('Model size:',modelsize)\
  model = whisper.load_model(modelsize) \
  return model\
\
def ML_inference_task():\
  text = model.transcribe(waveform.squeeze(), fp16=fp16flag, language='en')\
  return text\
  \
  \
from time import sleep, perf_counter\
\
# start the threads\
model = ML_model_load()\
\
\
start_time = perf_counter()\
ML_inference_task()\
end_time = perf_counter()\
\
\
latency = end_time - start_time\
process_speed = audio_len/latency\
\
print(f'It took \{end_time- start_time: 0.4f\} second(s) to complete.')\
print('Audio processing speed:',process_speed,'seconds of input audio per second')\
\
\
print('===========================================')}