Transcribe YouTube with Google Cloud Speech API (on Colab)

youtube_to_json.py

import json
from subprocess import check_output
from time import sleep
!pip -q install youtube-dl

from google.colab import auth
auth.authenticate_user()
!gcloud config set project kora-id
gcs = 'gs://co-lab/dhamma'  # use your own project, gcs

# main function
def convert_youtube(vid):
  download_audio(vid)
  oid = transcribe(vid)
  show_progress(oid)
  save_result(oid, vid)

""" download, convert to flac, then upload """
def download_audio(vid):
  !youtube-dl youtu.be/$vid --id --format m4a -q
  !ffmpeg -loglevel panic -y -i {vid}.m4a -ac 1 -ar 16000 {vid}.flac
  !gsutil -q cp {vid}.flac {gcs}/{vid}.flac   # upload to transcribe

def transcribe(vid):
  !gcloud ml speech recognize-long-running {gcs}/{vid}.flac \
      --language-code th --sample-rate 16000 \
      --include-word-time-offsets --async > {vid}.id
  return json.load(open(vid+'.id'))['name']

def show_progress(oid):
  cmd = 'gcloud ml speech operations describe '+oid
  while True:
    res = check_output(cmd.split()).decode()
    data = json.loads(res)
    pc = data['metadata'].get('progressPercent',0)
    if pc==100: break
    print('\r', pc, '% ', end='')
    for _ in range(60):
      print('.', end='')
      sleep(1)
  print('\r','100% finished')
  
def save_result(oid, vid):
  !gcloud ml speech operations wait {oid} > {vid}.json
  !gsutil -q rm {gcs}/{vid}.flac
  #!rm {vid}.m4a {vid}.flac {vid}.id