INNER CODE UNIT · Python

process_audio_emb

jdh-algo/JoyHallo · scripts/app.py:223

def process_audio_emb(audio_emb: torch.Tensor) -> torch.Tensor:
    """
    Process the audio embedding to concatenate with other tensors.

    Parameters:
        audio_emb (torch.Tensor): The audio embedding tensor to process.

    Returns:
        concatenated_tensors (List[torch.Tensor]): The concatenated tensor list.
    """
    concatenated_tensors = []

    for i in range(audio_emb.shape[0]):
        vectors_to_concat = [
            audio_emb[max(min(i + j, audio_emb.shape[0] - 1), 0)]for j in range(-2, 3)]
        concatenated_tensors.append(torch.stack(vectors_to_concat, dim=0))

    audio_emb = torch.stack(concatenated_tensors, dim=0)

View source record →