Institutional Repository of Vilnius University
FPGA panaudojimas garso atpažinimo uždaviniuose /
Abstract
dc:descriptionThis study investigates the usefulness of field-programmable gate arrays (FPGAs) for voice recognition tasks, exploiting their unique advantages of parallel processing capability, low latency and energy efficiency. FPGAs with scalable digital logic blocks and programmable interconnects are a favourable solution for accelerating computationally intensive tasks such as voice recognition. They allow algorithms to be implemented directly in hardware, facilitating efficient real-time processing. This study investigates the implementation of two different voice recognition tools: a convolutional neural network (CNN) and a digital signal processing (DSP)-based approach, using FPGAs. The CNN model is used because of it's excellent performance of CNNs in audio and speech processing tasks. To ensure compatibility with FPGAs, a model pruning approach is used to simplify the model and increase its computational efficiency. Conversely, the DSP-based tool uses signal processing techniques to extract relevant features from the input audio signal for subsequent classification. For the experimental setup, the Xilinx Zynq-7000 FPGA-based board PYNQ-Z1 is used due to its computational power, flexibility and compatibility with Python, TensorFlow and Keras. The main objective of this study is to develop and compare two FPGA-based voice recognition tools, one using convolutional neural networks and the other using a digital signal processing algorithm. A detailed comparative analysis of these tools will provide a comprehensive understanding of their strengths and weaknesses, thus pointing to future advances in FPGA-based voice recognition.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Prūselaitis, Simas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192847181&prefLang=en_US