Skip to main content

Speech-To-Text Filter

1. Download the Needed Models​

Download the required models from either the package provided by Capella or from GitHub:

2. Choose a Model​

  • If you downloaded the Capella provided "whisper-models" package, there will be 5 models available.
  • These models are multilingual and should work with different languages.

3. Configure FTC​

  • Open FTC and go to the filters list.
  • Select "Audio Speech Extraction".

4. Set Output File​

  • Click "Browse..." under "Output Filename" and select a destination for the output file.
  • Name and set the extension of the output file to anything you want.

5. Select a Model​

  • Under "Speech Extraction Model", choose a model to use. You can select any of the 5 packaged models.
    • Note: Smaller models are much faster but less accurate in extraction.
    • Note: Extraction quality also depends on the language.

6. Start Extraction​

  • Click "OK" to start the process.
  • The extracted output file should now be created in the selected location.

Troubleshooting & FAQs

Q: Why am I receiving this error? "Error: Configuration error: Internal Format ID: 20000"​

A: This error means your CPU is too old to use this feature. The CPU must support at least AVX2 to run this function.

Q: Are there any String Replacement variables that can be used?​

A: Yes, you can use a string replacement variable to make the output file name the same as the source name.

For example, when using the "Audio Speech Extraction" filter in WatchFolder, you can set the output path as:

C:\Users\Public\Documents\CapellaOutput\%sourceName%.mp4

This will automatically replace %sourceName% with the original file name.