Open Mega4alik opened 6 years ago
I tried to do iROVER with the predicted WER instead of confidence score, though I didn't get any improvement. but I didn't spend much time on it. And I might have done something wrong. I guess you can get better results with iROVER.
@shahabjld Did you implement it yourself or used available source code?
I implemented myself but again I didn't keep my code as it didn't improve my results.
On Sun, May 27, 2018, 10:14 Mega4alik notifications@github.com wrote:
@shahabjld https://github.com/shahabjld Did you implement it yourself or used available source code?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/hlt-mt/TranscRater/issues/5#issuecomment-392334068, or mute the thread https://github.com/notifications/unsubscribe-auth/AMuxDPAJm1yUy0lqpkNnpz5XJp-24TNRks5t2rTagaJpZM4UPOgR .
I believe that combining TranscRater with improved ROVER algorithm like iROVER (http://delivery.acm.org/10.1145/1620000/1614125/p65-hillard.pdf) can outperform TranscRater + ROVER combination.
Can @shahabjld or anyone else suggest well-performing ROVER improvement project with available source code?
Thanks in advance