SayLess: FPGA Voice Digit & Speaker Classifier

Document
Contributors
Degree granting institution: British Columbia Institute of Technology
Abstract
SayLess is a real-time voice digit and speaker recognition system implemented on an DE0-Nano- SoC FPGA using a NiosV (RISC-V) soft processor. The system captures a spoken digit (0–9) through a microphone, runs a complete machine learning inference pipeline in C on-chip, and displays both the recognized digit and the identified speaker on the 7-segment display - all without any cloud connectivity or external compute. The original project proposal described a fully hardware-only voice authentication system implemented in pure SystemVerilog, using Zero-Crossing Rate (ZCR) frequency matching against an enrolled passphrase. The final delivered system differs substantially from that proposal: the scope evolved from hardware-only authentication into a hybrid hardware/software classification system, incorporating a MATLAB-trained machine learning model (PCA + LDA) exported as a C header file and executed by the soft processor. Speaker identification was added as a bonus feature and was not required in the original proposal. This report documents what was originally planned, what changed and why, the complete architecture and algorithm, MATLAB training pipeline, accuracy results, and hardware implementation.

Refine your search

Number of pages
70 pages
Type
Form
Language
Course
ELEX 7660
Rights
This license enables reusers to copy and distribute the material in any medium or format in unadapted form only, for noncommercial purposes only, and only so long as attribution is given to the creator. CC BY-NC-ND includes the following elements: BY: credit must be given to the creator. NC: Only noncommercial uses of the work are permitted. ND: No derivatives or adaptations of the work are permitted. https://creativecommons.org/licenses/by-nc-nd/4.0/