FPGA Implementation for GMM-Based Speaker Identification

Abstract

In today's society, highly accurate personal identification systems are required. Passwords or pin numbers can be forgotten or forged and are no longer considered to offer a high level of security. The use of biological features, biometrics, is becoming widely accepted as the next level for security systems. Biometric-based speaker identification is a method of identifying persons from their voice. Speaker-specific characteristics exist in speech signals due to different speakers having different resonances of the vocal tract. These differences can be exploited by extracting feature vectors such as Mel-Frequency Cepstral Coefficients (MFCCs) from the speech signal. A well-known statistical modelling process, the Gaussian Mixture Model (GMM), then models the distribution of each speaker's MFCCs in a multidimensional acoustic space. The GMM-based speaker identification system has features that make it promising for hardware acceleration. This paper describes the hardware implementation for classification of a text-independent GMM-based speaker identification system. The aim was to produce a system that can perform simultaneous identification of large numbers of voice streams in real time. This has important potential applications in security and in automated call centre applications. A speedup factor of ninety was achieved compared to a software implementation on a standard PC.

MSP-MFCC: Energy-Efficient MFCC Feature Extraction Method With Mixed-Signal Processing Architecture for Wearable Speech Recognition Applications

“…Fully considering the arithmetic property, Jo et al [4] proposed an energy-efficient floating-point MFCC extraction architecture based on field-programmable gate array (FPGA) with the improvement of frequency transformation and optimization of bit-width. Some other works [11], [12] about efficient MFCC extraction are also proposed based on FPGA for low-cost speech recognition systems. In addition, efficient parallel implementation of MFCC feature extraction on graphics processing units (GPU) [13] and digital signal processor (DSP) [14] are presented showing faster extraction than CPU implementation.…”

“…It hardly decreases compared to the conventional implementation and proves that the difference is uninfluential to the recognition. Some conventional works [4], [11], [12] compare their recognition accuracy under different algorithms and even different datasets. It is unreasonable because the increase in accuracy may not be caused by feature improvement, but by better algorithms or datasets.…”

A Simplified Speaker Recognition System Based on FPGA Platform

“…Most existing speaker recognition accelerators are based on traditional algorithms, such as GMM or SVM (Table 7). The execution time of the accelerator according to MFCC and GMM proposed by EhKan in 2011 is 0.8ms per vector for speaker set of size 20 when the main frequency is 48MHz [22]. RamosLara proposed an accelerator on the basis of MFCC and SVM.…”

GPU Implementation of Pairwise Gaussian Mixture Models for Multi-Modal Gene Co-Expression Networks

“…Gaussian mixture models are used for a wide variety of tasks, including background subtraction [15] and speaker verification [16], [17]. Additionally, several GMM implementations for these domains have been published for GPUs [18], [19] and FPGAs [20], [21], as well as general-purpose GMM implementations for GPUs [22] and FPGAs [23]. Arguably the most similar work to ours is [19], in which the authors accelerated a GMM-based background subtraction algorithm by processing each pixel-wise GMM in a separate thread, in the same way that the KINCv3 GMM kernel devotes a thread to each gene pair.…”