Audio

Experimental

What is it?

Audio models cover speech recognition (ASR, e.g. Whisper-style), speech synthesis (TTS) and audio understanding. The badge appears for repositories tagged with audio/speech tasks.

When should I use it?

Transcribing meetings or voice notes fully locally.
Producing spoken output without a cloud TTS.

Prerequisites

An audio GGUF — filter the catalog by Audio.
Runtime support for that audio architecture. Experimental Support varies per model family; check the model detail page and status.

Minimal working example

Planned Audio transcription/synthesis endpoints are not part of the local API yet. The intended surface: audio input accepted on chat-style endpoints for ASR models, and a dedicated audio generation route for TTS. This section documents intent only — do not expect a working call today.

Configuration/options

Sample rate and format conversion are the caller's responsibility in the planned interface.
Audio models often pair a decoder with the main GGUF; both files appear in the model's file list.

Limitations

Model/runtime dependent — far more than text: audio needs architecture-specific runtime support.
Long recordings consume context proportional to audio length.
Badge is inferred from Hub tags; not all tagged repos are usable locally.

Troubleshooting

Model loads but audio requests fail → runtime lacks support for the architecture; see Runtime problems.