Aqib Ahmad
FAST
· 2022
·
i16 - 0019
Email
—
Phone
—
LinkedIn
—
GitHub
—
Academic
Program
BSCS
CGPA
—
Year
2022
Education
SEECS
Address
—
DOB
—
Career
Current role
—
Target role
—
Skills
Python, PyTorch, Google Colab, Deep CNNs, Facial Recognition, Voice Matching
Verbatim text
The exact text the LLM saw on the page (or the booklet text from the old import).
This is what powers semantic search.
Speech2Face Speech2Face is an R&D-based project that aims to learn the correspondence between facial and vocal features of humans which is then used to match a voice sample to a particular facial image. The problem complexity is such that sound cannot be directly mapped to a facial image using a single neural network. Hence, we employ the use of two Deep CNNs (Convolutional Neural Networks) for the purpose. The details of the models are as follows: Sound to Face-Vector Model: This model converts a given soundwave into a facial recognition vector. This is the same vector that is the output of face recognition systems like Facenet. It contains facial features of the person. Face-Vector to Face-Image Model: This model takes the same facial features vector that is output of the 1st model, and generates a front-facing neutral expression image of the person. Hence, it is most complex of the two models. It gives a state-of-the-art level of image compression. \
Provenance
Source file: Graduate Directory FAST School of Computing 2020 (Final Complete) (1).pdfFrom job #23 page 223
Created: 1778226103