AutoSub is a CLI application to generate subtitle file (.srt) for any video file using Mozilla DeepSpeech. I use the DeepSpeech Python API to run inference on audio segments and pyAudioAnalysis to split the initial audio on silent segments, producing multiple small files.

⭐ Featured in DeepSpeech Examples by Mozilla


In the age of OTT platforms, there are still some who prefer to download movies/videos from YouTube/Facebook or even torrents rather than stream. I am one of them and on one such occasion, I couldn't find the subtitle file for a particular movie I had downloaded. Then the idea for AutoSub struck me and since I had worked with DeepSpeech previously, I decided to use it.


Clone the repo. All further steps should be performed while in the AutoSub/ directory

$ git clone
$ cd AutoSub

Create a pip virtual environment to install the required packages

$ python3 -m venv sub
$ source sub/bin/activate
$ pip3 install -r requirements.txt

Download the model and scorer files from DeepSpeech repo. The scorer file is optional, but it greatly improves inference results.

# Model file (~190 MB)
$ wget
# Scorer file (~950 MB)
$ wget

Create two folders audio/ and output/ to store audio segments and final SRT file

$ mkdir audio output

Install FFMPEG. If you're running Ubuntu, this should work fine.

$ sudo apt-get install ffmpeg
$ ffmpeg -version               # I'm running 4.1.4

[OPTIONAL] If you would like the subtitles to be generated faster, you can use the GPU package instead. Make sure to install the appropriate CUDA version.

$ source sub/bin/activate
$ pip3 install deepspeech-gpu


  • Installation using Docker is pretty straight-forward. The model build-arg configures which model and scorer versions to use. You can manually edit them to point to other model files easily.
$ docker build --build-arg model=0.9.3 -t ds-stt .
$ docker run ds-stt --file video.mp4
$ docker cp <container-name>:/output/ /<your-local-dir>/
  • Make sure to use container name while copying to local.

How-to example

  • Make sure the model and scorer files are in the root directory. They are automatically loaded
  • After following the installation instructions, you can run autosub/ as given below. The --file argument is the video file for which SRT file is to be generated
$ python3 autosub/ --file ~/movie.mp4
  • After the script finishes, the SRT file is saved in output/
  • Open the video file and add this SRT file as a subtitle, or you can just drag and drop in VLC.
  • WEB VTT Output (Credits - @DerrickGibbs1): Output VTT file including cue points for individual words. Nearly identical to VTT file downloaded from YouTube with youtube_dl.
$ python3 autosub/ --file ~/movie.mp4 -vtt

How it works

Mozilla DeepSpeech is an amazing open-source speech-to-text engine with support for fine-tuning using custom datasets, external language models, exporting memory-mapped models and a lot more. You should definitely check it out for STT tasks. So, when you first run the script, I use FFMPEG to extract the audio from the video and save it in audio/. By default DeepSpeech is configured to accept 16kHz audio samples for inference, hence while extracting I make FFMPEG use 16kHz sampling rate.

Then, I use pyAudioAnalysis for silence removal - which basically takes the large audio file initially extracted, and splits it wherever silent regions are encountered, resulting in smaller audio segments which are much easier to process. I haven't used the whole library, instead I've integrated parts of it in autosub/ and autosub/ All these audio files are stored in audio/. Then for each audio segment, I perform DeepSpeech inference on it, and write the inferred text in a SRT file. After all files are processed, the final SRT file is stored in output/.

When I tested the script on my laptop, it took about 40 minutes to generate the SRT file for a 70 minutes video file. My config is an i5 dual-core @ 2.5 Ghz and 8 gigs of RAM. Ideally, the whole process shouldn't take more than 60% of the duration of original video file.


  • Pre-process inferred text before writing to file (prettify)
  • Add progress bar to extract_audio()
  • GUI support (?)


GitHub - abhirooptalasila/AutoSub: A CLI application to generate subtitle file for any video using Mozilla DeepSpeech
A CLI application to generate subtitle file for any video using Mozilla DeepSpeech - GitHub - abhirooptalasila/AutoSub: A CLI application to generate subtitle file for any video using Mozilla DeepS...