video-auto-caption

Thie project contains Nodejs scripts which generate captions/subtitles for videos. It is done as a winning submission for Hack&Roll 2018 but it is by far from production ready. The only working file is captioner.js which only allows generating captions for videos under 1 minute. This is a limitation set by Google Speech API Synchronous Speech Recognition.

Prerequisites

You will need to create a billing account on Google Cloud Platform in order to use the Speech API but accessing the API will be free up to a point. Check on https://cloud.google.com/speech/pricing

Go to https://cloud.google.com/speech/docs/quickstart and click on the "Set up a project" button. A JSON file containing the service account private key will be downloaded.

Set the environment variable for GOOGLE_APPLICATION_CREDENTIALS as the path string to the JSON file (https://cloud.google.com/docs/authentication/getting-started#setting_the_environment_variable).

Install ffmpeg which is a dependency to node-fluent-ffmpeg. Make sure you have PATH variable set for the install directory.

Getting Started

Installing

npm install

Test run

node captioner.js fox.mp4

Running

Currently, the only completely working file is captioner.js.

You can run it by

node captioner.js input.mp4 output.mp4

• input.mp4 is be the file name of your source video you want to have it captioned (which should be in the project directory),

• output.mp4 is optionally the file name of the output video.

• If output.mp4 is omitted, the output file name defaults to input-captioned.mp4 where input is the source video file name.

Keep in mind that the Speech API needs very clear speech audio. As of the duration of the hackathon, I estimated about 70% accuracy on speech recognition with poor recording environment.

captioner.js

Here I will briefly explain what the code does.

1. It uses ffmpeg to extract out the audio from the video file and convert to single channel (limitation of Speech API) and save it to sample.flac.
2. Then it generates a sample.srt file (which is a subtitle file format) by sending the flac file over to the Speech API for it to detect words. The API will then return a list of words along with the corresponding timestamp of each word. It then chunks up words which are within 3 seconds apart and save it as a phrase into the srt file.
3. It again uses ffmpeg to combine the source video and the subtitle into a single file.

Note that the output video will have different codec from the source video. I still have yet to completely understand how ffmpeg works but hey its from a hackathon :p

4. Finally the code deletes the sample.flac file but I keep the sample.srt file. This way you can manually modify the srt file to correct wrongly identified words and use the ffmpeg command line to combine the srt file with a video file like ffmpeg -i infile.mp4 -i infile.srt -c copy -c:s mov_text outfile.mp4 source.

Built With

• Google Speech API - Speech Recognition
• node-fluent-ffmpeg - Fluent interface for FFMPEG
• FFMPEG - Media manipulation

Contributing

I will be honest, this project is littered with bugs and bad practices but hey its from a hackathon :p So if you can understand my code and want to contribute, feel free to do so!

Authors

• David Choo - Quick 24 hour hack

See also the list of non existing contributors who participated in this project.

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• The amazing Hack&Roll 2018 for giving me an opportunity to learn something new over the weekend.