lowerquality / gentle

gentle forced aligner
https://lowerquality.com/gentle/
MIT License
1.45k stars 295 forks source link

Server running on the Docker image is slow #329

Open Akeyo21 opened 8 months ago

Akeyo21 commented 8 months ago

Issue

Whenever we try to run forced alignment for audios when running the Docker image locally the response takes really long. For a 14 second audio, it takes roughly 5 minutes to get the response. However, when we run forced alignment on the Mac application for the same audio, it takes less than 30 seconds.

Was this how it was intended to work, have the docker image run slower than the Mac application?

The docker image is run on a Mac M2 Pro, with 32GB RAM on OS Ventura 13.4

How to reproduce

  1. Pull the image by running the command:

    docker pull lowerquality/gentle:latest
  2. Run the image by running the command:

    docker run -P lowerquality/gentle
  3. Test the server by either: a. sending a HTTP request either via Postman or curl. For curl, you can run the command:

        curl -F "audio=@audio.mp3" -F "transcript=@words.txt" "http://localhost:8765/transcriptions?async=false"

    b. Running the application locally, by typing this in the browser: http://localhost:[portnumber]/ and try and align audio

niranjanakella commented 3 months ago

Yes @Akeyo21, experienced the same because the docker resources aren't enough to infer this system properly. Hence running the native Mac application they provide yielded good results for me.

ctstewart commented 2 months ago

@niranjanakella Unfortunately the native Mac application doesn't seem to work anymore.