Generate subtitles for video file

Using the paid Google Cloud Speech-To-Text API

This program requires a Google account and an API key: Create project on Google Cloud

$ ./srtgen.py 
usage
 srtgen.py --apikey path/to/keyfile.json path/to/input-video.mp4
environment variables
 GOOGLE_APPLICATION_CREDENTIALS=path/to/keyfile.json srtgen.py path/to/input-video.mp4
keyfile
 This program requires a Google account and an API key
 https://console.cloud.google.com/projectcreate

subtitle is written to stdout and output/xxxxxx-input-video.mp4/output_file.srt
where xxxxxx is the sha1 hash of the input video file

temporary files are stored in output/xxxxxx-input-video.mp4/

• workaround size limit in google API
  • no need for Google Cloud Storage = gs protocol
  • duration is limited to 60 seconds
  • file size is limited to 10485760 bytes

• ffmpeg
• python
  • pydub
  • google.cloud.speech
    • API key
    • pricing
      • speech recognition needs lots of space and time = there is no free lunch
      • https://cloud.google.com/speech-to-text/pricing#pricing_table
        • first hour is free
          • TODO one hour per month or one hour per google account?
        • Speech Recognition without Data Logging: 0ドル.006 / 15 seconds = 0ドル.024 / 1 minute = about 1ドル.50 / 1 hour
        • Speech Recognition with Data Logging: 0ドル.004 / 15 seconds = 0ドル.016 / 1 minute = about 1ドル.00 / 1 hour
        • Data Logging = feedback of manually corrected text to improve quality of service
          • TODO implement upload of corrected text
    • TODO Automatic punctuation

• https://github.com/plutowang/generate-video-subtitle
• https://cloud.google.com/speech-to-text/docs/basics

• edit subtitles
  • aegisub
  • gaupol
  • subtitleeditor
• translate subtitles
  • translatesubtitles.co

• https://github.com/BingLingGroup/autosub
  • online speech recognition
    • Google
    • Xfyun
    • Baidu
• https://github.com/abhirooptalasila/AutoSub
  • offline speech recognition
    • Mozilla DeepSpeech
    • lower quality than google speech
    • limited by user hardware: space, time, cpu instruction set (binaries dont run on weak cpus)
• https://github.com/topics/subtitles-generator
  • https://github.com/nestyme/Subtitles-generator

• use speech_recognition module, so srtgen can use multiple backend services
  • support offline speech recognition
    • mozilla deepspeech
    • vosk
    • Picovoice https://picovoice.ai/docs/picovoice/
  • we need a service that returns timestamps for every word
    • google cloud speeech: enable_word_time_offsets=True
    • alternative: synchronize words and audio waveform
      • https://github.com/otsaloma?tab=stars&q=subtitle
        • https://github.com/smacke/ffsubsync Automagically synchronize subtitles with video.
        • https://github.com/kaegi/alass "Automatic Language-Agnostic Subtitle Synchronization"
• hybrid of offline and online speech recognition
  • deepspeech for offline speech recognition
  • google for online speech recognition
  • can deepspeech return confidence values?
  • run deepspeech with different models? (and manually select the best result?)
• automatic postprocessing
  • reduce manual work
  • split long sentences
  • merge short sentences