Skip to main content

When to Use Custom Models

Scriptix gives you two ways to handle domain-specific vocabulary: custom models (which improve transcription) and glossaries (which control translation). They solve different problems at different stages.

Quick Decision​

You have...Use...
Audio files and matching transcripts (.vtt, .srt, .txt), and want better transcriptionCustom model
A short list of terms you want translated a specific wayGlossary
Both needsBoth (custom model for transcription, glossary for translation)

Custom Models​

Good for: ongoing work in a specialised domain (medical, legal, product names, dialect) where you have or can produce paired audio + transcripts.

Requires:

  • A trainable base language (only trainable languages appear in the create dropdown).
  • Audio in .wav, .mp3, .m4a, .flac and/or transcripts in .vtt, .srt, .txt.
  • Training credits on your organisation's plan.

Trade-offs: training takes time and consumes credits. You can't change the base language after creating a model - only the name.

If you have audio but no timestamped transcripts, the in-app guide and the detail-page banner point you at Force Alignment, which adds timestamps to plain-text transcripts so you can use them as training data. It's available from the Create button on the dashboard.

Glossaries​

Good for: small, exact term replacements you can write down - product names, acronyms, source to target translations of specific terms.

Requires: a list of term pairs, plus a source and target language. No training, no audio.

Trade-offs: glossaries apply during translation, not transcription. They keep chosen terms consistent when a transcript is translated, but they don't change how the recogniser hears speech.

Using Both​

A custom model improves how audio is transcribed; a glossary keeps specific terms consistent when that transcript is later translated. Use a custom model to get an accurate transcript, and a glossary to control the terminology in its translations.

Next Steps​