Different embedding dimensions

We are passing videos of different durations through the model and we are getting audio embedding of different sizes. Is this normal ? 
We are looking forward to use the audio embeddings generated from your model in our network, but since the embedding size is variable, we cant move forward with it. Is there any solution for it? 
Thanks in advance.