Voice activity detection system for smart earphones

Abstract

This paper presents a real-time voice activity detection (VAD) algorithm implemented in a miniature Digital Signal Processor (DSP) for in-ear listening devices such as earphones or headphones. This system allows consumers to hear external speech signals such as public announcements or oral communication while listening to music without removing their listening devices. The proposed algorithm uses two normalized energy features that compare the energy in the frequency region containing speech information with the frequency regions typically containing noise. The extraction of the normalized features represents the key of the proposed VAD since it eliminates the need for a signal-to-noise ratio (SNR) estimator. The VAD's decision is made using two threshold comparison rules computed from the normalized features and a hangover scheme triggered after a given number of observations. The algorithm parameters, namely the frequency regions' boundaries, number of observations, two decision thresholds and hangover's duration, have been optimized offline using a genetic algorithm. The performance of the proposed VAD is compared to a benchmark algorithm in four noise environments and three SNRs. Results show that the average false positive rate (FPR) of the proposed algorithm is 4.2% and the average true positive rate (TPR) is 91.4 % compared to the benchmark algorithm which has a FPR average of 29.9 % and a TPR average of 79.0 %. The proposed VAD is implemented in hardware to validate its reliability and complexity 1 .Index Terms -Smart earphones, voice activity detection, energy based feature, real-time algorithm, digital signal processor. N. Lezzoum is with the

Hardware Implementations for Voice Activity Detection: Trends, Challenges and Outlook

“…Sampling frequency and sample width are important design parameters for a digital VAD but either or both the parameters are not provided in [5], [15], [20]. Mixed-signal designs presented in [9], [19], [22]- [24] do not provide complete information about the bandwidth, sample frequency or sample width.…”

“…Covering the last 12 years (Jan 2010 -Dec 2021), we found several hardware implementations of VAD (either mea- sured or simulated at the device level using foundry PDKs) [5]- [25]. This includes digital ( [5], [7], [16], [18], [20], [21]), analog ( [8], [10]), and mixed-signal ( [6], [9], [17], [19], [22]- [25]) designs along with some implementations on Field Programmable Gate Array (FPGA) ( [13], [14]), Field Programmable Analog Array (FPAA) ( [12]), Digital Signal Processor (DSP) ( [15]), and Neuromorphic platforms (e.g. Intel Loihi [26]) ( [11]), see Fig.…”

A Convolutional Neural Network Smartphone App for Real-Time Voice Activity Detection

“…As far as real-time VADs are concerned, in [19], Lezzoum et al utilized normalized energy features along with a thresholding technique. The real-time VAD developed by Sehgal et al in [2] was implemented to run on smartphones as an app using the features developed in [1].…”

Assessment of sound source localization of an intra-aural audio wearable device for audio augmented reality applications

“…These AWDs are capable of protecting the ear from background noise while simultaneously transmitting warning signals to the wearer's ear [3] and enabling face-to-face communication by detecting and transmitting enhanced speech signals to the protected ear [4].…”