Open vkeep opened 1 month ago
Hi vkeep,
Thank you for your question.
The model structure we referred to is based on the short version of MP-SENet accepted by Interspeech 2023. Since the long version of MP-SENet has not yet released the official code, we chose to compare with the Interspeech 2023 version of MP-SENet to prevent any mismatches in re-implementation details.
Additionally, to the best of our knowledge, the long version of the MP-SENet model has some differences in design. Therefore, we compared the short version’s conformer with the implemented Mamba in Table 2.
We will update the result scores and further investigate our existing model once the long version of MP-SENet is officially released.
Best regards, Roy Chao
from latest MP-SENet long article https://arxiv.org/abs/2308.08926, the PESQ results is 3.6 on the voicebank +demand dataset, your paper instead of conformer by mamba ,the results didn't increase, will you update the MP-SENet comparison results?