1. 1500 USD
2. 500 USD
3. 250 USD
Subject: Speaker Identification by Deep Learning
Description: Among 2K speaker voices (wav files), finding the new given voice’s speaker. Each train voice data will be at least 90 seconds and test voices will be given as 3sec. Handling voice record channel variation will be required. Solution should be applicable for 1M voices well.
Website: www.evam.com
Link for registration: by mail to [email protected]
Tools: Deep learning model will be coded in Python preferable by using one of Pytorch, Keras, Tensorflow or coder can use other Python deep learning packages.
Deadline for submission: 14th of June 2020 17:00 (GMT)
● After the submission, we will give 50 sample voices with 3 secs duration and we will ask attendees to find the speakers of those voices with respect to the voice files used for training the algorithm. For each voice similarity ratios will be listed with top 20 similar speakers and each of them needs to be scored with similarity scores.
● well documented and commented code
● performance of the algorithm in terms of time and accuracy
● submission duration, earlier submissions get higher marks
● runnable, well-commented code with start script
● instructions about how to start the code, train the training voice files, and prediction
● contact details of the person or the group
● a presentation that answers below questions :
o feature extraction process of the algorithm
o model training and evaluation
o how the embedded vector(s) stored, if any
o how the enrollment is made
o how the test voice is evaluated
For questions and submissions: Can Alhas, [email protected]
