For A Voice Embedding Better With or Without trailing Silence?

[EthanEpp]

Working with the voice embedding, and if I have a captured brief sample of a user's speech (a few seconds) is it better to have the audio used to generate the embedding be trimmed to the point of starting right at the start of speech, or is a half to full second of silence at prior to speech ideal? Same question for at the end of speech.

I'd be inclined towards as little silence as possible without risking clipping the target speech, but wondering if there is any input here.

Thanks!

[hbredin]

I confirm you'll get better results by removing non-speech.