Models

Models

This is where you decide the core transcription engine Voiscribd will run.

Model Sources

Use model filters to switch between:

• Local
• Cloud
• Custom

Default Model

Every model card can be set as default for transcription.

Your selected default is used unless a Chain overrides it.

Language Behavior

Language options depend on model capabilities:

• multilingual models: Auto or fixed language
• English-only models: fixed English behavior
• some cloud transcription families auto-detect language in their own pipeline

Language Selection

Use fixed language when consistency matters more than flexibility.

Import Local Models

Local tab includes Import Local Model for .bin Whisper-style model files.

Add Cloud Model

Cloud tab includes Add Cloud Model with:

• provider
• display name
• async model name
• optional description
• multilingual toggle
• optional live transcription enablement (for supported providers)
• optional dedicated realtime model name

Use cloud models when you prioritize convenience and managed infrastructure.

Add Custom Model

Custom tab includes Add Model with:

• display name
• transcription API endpoint
• API key
• model name
• multilingual toggle

Use this for OpenAI-compatible transcription backends not bundled in default cloud list.

Practical Selection Strategy

1. choose one stable default model
2. verify language behavior on your real audio
3. use Chains to override model per context instead of changing global default repeatedly

Common Mistakes

Quality is inconsistent across apps

Usually a Chain override is active. Check active chain first.

Wrong language detection

Set explicit language for that model/context and retest.