Offline subtitles experiment

Speech-to-text

I am working on online courses, and subtitles are a good idea, so I wondered how well it worked.

After some searching, many suggests using the google api for that and a lot of paid services. E.g. Pytranssciber uses google, and looking at the google pricing model you get 1h for free per month, which is not enough for my use.

There are some opensource software like deepspeech and wav2letter. The latter is a facebook project and “it has been moved to flashlight”. Also CMUSphinx (and pocketsphinx) could be relevant.

Since I want subtitles not just the text, I found AutoSub which uses deepspeech.

Testing it

I have mad a gitlab project with the Vagrantfile to get started. See the readme file for details.

There is a dependency on the vbguest plugin, this was needed to get two-way sync running using Virtualbox. Also, I needed to use contrib-buster/debian64 basebox which includes the vboxfs. On my dev machine, I use KVM+NFS, so there two-way “sync” was already working (see a previous post about that.

The repo is on gitlab.

If the Vagrant part is interesting, I have made a “getting started” free course about it.