To use SpeechT5 with the pipeline, you'll need a speaker embedding.