Open arpita739 opened 3 years ago
Haven't till now but may be I will write article on this.
You can follow me on medium😊
Okay
On Tue, Jan 12, 2021, 2:40 PM Vikas Kumar Ojha notifications@github.com wrote:
You can follow me on medium😊
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/VikasOjha666/Speaker-Identification-One-Shot-Learning/issues/1#issuecomment-758515696, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJWPRA65VY7NG7N4CMWBFHTSZQGYNANCNFSM4V63764Q .
Hey this model u shared in the GitHub. Did u downloaded the whole training dataset and then run the model or just did it for some and then run the model with that trained data?
On Tue, Jan 12, 2021, 2:49 PM arpita halder arpitahalder739@gmail.com wrote:
Okay
On Tue, Jan 12, 2021, 2:40 PM Vikas Kumar Ojha notifications@github.com wrote:
You can follow me on medium😊
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/VikasOjha666/Speaker-Identification-One-Shot-Learning/issues/1#issuecomment-758515696, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJWPRA65VY7NG7N4CMWBFHTSZQGYNANCNFSM4V63764Q .
Actually I trained this model on Google Colaboratory.Hence I downloaded the whole dataset.However considering the number of batch in epoch the training was not sufficient as I was just trying to prove the concept for later use in my future project.You can train this model further to achieve better performance.In case of problem you can contact me at evilangel1998666@gmail.com or here too.I will be happy to help.
Hi @VikasOjha666, great job! Just curious, was this model able to achieve real-life acceptable accuracy? And how will speaker enrolment be done?
@kelvine95 My aim behind this project was just to figure out a method to perform speech comparison. I had trained this with a small subset of librispeech dataset hence it wouldn't have acceptable real-life accuracy. But by training the model in a similar way with a bit of tuning in neural architecture and a bit of data augmentation we can achieve real-life acceptable accuracy. In my case there were limitations like computation as well as time as I was in college at the time I created this project.
Have you prepared any write up for this?