.

[stevetsim]

.

[jongwook]

Does your audio have multiple languages and code-switch between them? The model is trained to only do transcription with task="translate" but it may get confused when there are multiple languages.