Closed AbdallahQoutbAli closed 4 years ago
Hi, thank your for your interest in the repository.
What data did you use to train the network and how similar is it to the data you are trying to predict? Unfortunately the network is not very good at generalizing. These are my results on the AMI and ICSI meeting corpora: with
AAA
-> Trained on AMI and tested on AMIIII
-> Trained on ICSI and tested on ICSIAAI
-> Trained on AMI and tested on ICSIYou can see, that the results on the cross-corpus test (AAI
) are significantly worse. Do not expect too much from the network, as this was mainly an experiment.
More details are described in chapter 5 (Experiments) and chapter 7 (Summary and Conclusion) of my bachelor's thesis: https://github.com/Bastian/bachelors-thesis/releases/download/submission/thesis-colored-links.pdf
I hope this helps! :slightly_smiling_face:
Thank very much for replay I'm training model using your files in data folder after 80 epoch give me Good output in Your file predict. but not Good in my text Link : https://drive.google.com/file/d/1NPI7gjRVzirHjvgv9fX1QIC3uWCmoJ85/view?usp=sharing the text content interview between to person, the ouptut using model speech to text in my graduation project Now I will read chapters ,training in AMI data, and test text again
Hello Can you upload pre-traind model ?
Sorry for the late response, I somehow missed your comment. If you still need it, here you go: Here's my output folder after training on the AMI dataset: https://drive.google.com/drive/folders/1XD694hafdMFirCRitwMhkVFchxFJvCGq?usp=sharing :)
Hellow i'm tring to add my meeting text to but the result not good in file predit-predctions.txt before add text must run Meeting-Parser. ?