Automatic Voice Identification after Speech Resynthesis using PPG

Acoustic individual discrimination has been demonstrated for a wide range of animal taxa. However, there has been far less scientific effort to demonstrate the effectiveness of automatic individual identification, which could greatly facilitate research, especially when data are collected via an acoustic localization system (ALS). In this study, we examine the accuracy of acoustic caller recognition in long calls (LCs) emitted by Bornean male orangutans (Pongo pygmaeus wurmbii) derived from two data-sets: the first consists of high-quality recordings taken during individual focal follows (N = 224 LCs by 14 males) and the second consists of LC recordings with variable microphone-caller distances stemming from ALS (N = 123 LCs by 10 males). The LC is a long-distance vocalization. We therefore expect that even the low-quality test-set should yield caller recognition results significantly better than by chance. Automatic individual identification was accomplished using software originally developed for human speaker recognition (i.e. the MSR identity toolbox). We obtained a 93.3% correct identification rate with high-quality recordings, and 72.23% with recordings stemming from the ALS with variable microphone-caller distances (20–420 m). These results show that automatic individual identification is possible even though the accuracy declines compared with the results of high-quality recordings due to severe signal degradations (e.g. sound attenuation, environmental noise contamination, and echo interference) with increasing distance. We therefore suggest that acoustic individual identification with speaker recognition software can be a valuable tool to apply to data obtained through an ALS, thereby facilitating field research on vocal communication.

Speaker Identification systems have become area of intense research in recent years due to its wide range of applications in voice matching, biometric identification, mobile access security, health care management and transportation. The major challenge in speaker identification is how to provide robust and computationally efficient features that accurately captures speaker's unique identity for higher generalized identification. This paper presents hierarchical classification approach for speaker identification using robust time domain features. In hierarchical classification approach, the first level classifier identifies the gender voice (i.e. male or female voice). In addition, the second level classifier identifies the specific speaker voice. The experiments were conducted using the speech data collected from ten subjects including male and female to ensure the generalized model evaluations. Several time domain features were extracted from the collected dataset that were proven highly discriminative for gender identification and specific speaker identification. The proposed method helps to reduce computational time with increase speaker recognition rate. The experimental results obtained highest accuracy of 96.9% using random forest classifier for gender identification. Moreover, for specific speaker identification the highest accuracy of 78% was observed in male speaker identification and 88.7% accuracy was obtained in female speaker identification using random forest classifier. The encouraging results of our experiments show the impact of hierarchical speaker identification using carefully selected time domain features.

Biometric template protection of speech signals and information hiding in speech signals are two challenging issues. To resolve such limitations and increase the level of security, our objective is to build multi-level security systems based on speech signals. So, speech watermarking is used simultaneously with automatic speaker identification. The speech watermarking is performed to embed images into the speech signals that are used for speaker identification. The watermark is extracted for authentication, and then the effect of watermark removal on the performance of the speaker identification system in the presence of degradations is studied. This paper presents an approach for speech watermarking based on empirical mode decomposition (EMD) in different transform domains and singular value decomposition (SVD). The speech signal is decomposed in different transform domains with EMD to yield zero-mean components called intrinsic mode functions (IMFs). The watermark is inserted into one of these IMF components with SVD. A comparison between different transform domains for implementing the proposed watermarking scheme on different IMFs is presented. The log-likelihood ratio (LLR), correlation coefficient (Cr), signal-to-noise ratio (SNR), and spectral distortion (SD) are used as metrics for the comparison. According to the simulation results, we find that the watermark embedding in the discrete sine transform domain provides higher SNR and Cr values and lower SD and LLR values. The proposed approach is robust to different attacks.

Hands-free operation of speech processing equipment is sometimes desired so that the user is unencumbered by hand-held or body-worn microphones. This paper explores the use of array microphones to capture speech under adverse acoustic conditions, and provide input to a system for automatic speaker identification. The system is evaluated using reverberated speech signals, generated by a computer model of room acoustics and transduced by different simulated microphone-arrays. For comparison, the system is also evaluated using close-talking microphone input. It is found that 2-D matched-filter microphone arrays are capable of producing high speaker identification scores in a hostile acoustic environment, such as multipath distortion and competing noise sources. The paper also explores the influence of vector quantization techniques, codebook size, and order of cepstrum coefficients on the performance of the speaker identification system. >

Automatic speaker identification techniques are widely used nowadays in forensic applications, but its accuracy harshly drops when the voice of the speaker of interest is immersed in a recording containing more than one voice, common situation of investigations where the targets voice are obtained through ambient recordings. In forensic applications where microphones are hidden, such interferent sound sources in recordings are common and they degrade severely the performance of speaker identification techniques. In this paper, we propose a method to mitigate this problem by spatially separating the voice of each speaker using a Blind Source Separation technique called Convolutive Independent Component Analysis, and then applying the separated speech signals to a speaker identification system based on Mel Frequency Cepstral Coefficients and Gaussian Mixture Models. For identifying more than one speaker, the proposed system has a better accuracy than the state-of-the-art solutions.

This paper concerns the problem of the effect of emotional change on humans and machines for speaker identification. A contrasting experiment is carried out between Automatic Speaker Identification (ASI) system (applying GMM-UBM and Emotional Factor Analysis (EFA) algorithm) and aural system on emotional speech corpus MASC. The experimental result is similar to that in channel-mismatched condition, i.e. the ASI system is much better than the single listener, especially when emotion compensation algorithm EFA is applied. Meanwhile, fusion of multiple listeners can significantly improve the aural system performance by 23.86% and make it outperform the ASI system.

Speaker identification answers the fundamental question Who is speakingƒ The identification technology enables various downstream applications to provide a personalized experience. Both the prevalent i-vector based solutions and the state-of-the-art deep learning solutions usually treat all users equally, with no distinctions between new users and existing users, during the training process. We notice that a good many new users start with limited labeled training data, which often results in inferior predicting performance of recognizing users’ voices. To alleviate the disadvantage caused by training data deficiency, we propose a Mixture Density Network- based Meta-Learning method (MDNML) for speaker identification. MDNML emphasizes the expeditious process of learning to recognize new users where each has only a few seconds of labeled data.We conduct experiments on the LibriSpeech dataset and compare MDNML with four state-of-the-art baseline methods. The results conclude that MDNML achieves higher accuracy in recognizing new users with limited labeled utterances than all baseline methods. Our proposed solution significantly expedites the learning by transferring the knowledge learned from the existing user base through gradient-based meta-learning. We consider our work to be a steppingstone for more sophisticated meta-learning frameworks for accelerating voice recognition. Furthermore, we discuss a strategy for enhancing the accuracy by incorporating the notion of household-based acoustic profiles with MDNML.

A two stage fuzzy decision classifier for speaker identification

A novel approach for Automatic Speaker Identification (ASI) employing Waveform based signal representation in multiple domains is presented. The proposed approach involves two stages, namely, the encoding stage, and the decoding stage. During the encoding stage (training mode), mixed transform coding, in conjunction with split vector Quantization (MTSVQ) is employed to form representative codebooks for each speaker. During the decoding stage (running mode), the vectors that best represent the unknown input vector are selected to represent the speech vectors. A normalised matching accuracy measure is developed to evaluate the proposed algorithm's performance. The resulting technique is consistently found to obtain enhanced ASI accuracy in comparison with the earlier approaches as vector quantization employing single transform domains.

Results of an experimental study and the optimization of features for a conventional vector-quantization codebook-based automatic speaker identification (ASI) system are presented. Standard LPC (linear predictive coding) and a perceptually weighted feature termed PLP (perceptually based linear prediction) are compared using appropriate distance measures, namely, the log-likelihood, and three cepstral variants: constant weighting, the robot-power-sum, and the inverse variance. PLP features combined with a weighted cepstral measure are found to be consistently the best in a number of different digit-independent ASI experiments. Results support the hypothesis that the higher orders of PLP (>5) contain significant speaker-specific information, with ASI performance improving rapidly up to order 8, and then far more slowly yet consistently up to order 16. A similar pattern is seen for codebook size, with fast improvements up to size 64, with more gradual gains thereafter. >

This research paper will present an automatic Speaker Identification System using MFCC (Mel-Frequency Cepstral Coefficients) and BPNN (Back Propagation Neural Network). The objective of this work is to classify 20 speakers’ pattern and to identify each registered speaker correctly while testing with new input speech without any false identification. MFCC is used for the extraction of speech features from each speaker and BPNN is used for identification of the test speaker.  The developed classifier model is tested with both registered and unregistered speakers and found that it successfully identifies all the registered speakers correctly and reject the intruder speakers. Scaled conjugate gradient training function is used for training the BPNN. A speech database consisting of 20 speakers is created from a group of 10 male and 10 female speakers with the same sentence spoken twice. The classification accuracy rate obtained from the classification is 92.1% and the correct identification rate obtained is 100%. Matlab simulation tool is used in this work

While few individuals would argue that vocal cues can signal a person’s identity, it is difficult to specify exactly which parameter(s) provide the most salient information for speaker identification. Previous literature has suggested that speaking fundamental frequency, long-term spectra, vowel formant frequencies, and speech tempo can provide speaker-specific information. However, investigations focused on automatic speaker identification have provided less than satisfactory results. These findings could be related to how each acoustic parameter is measured or, more probably, to the idea that these acoustic parameters interact in specific ways that may be more obvious in the perceptual realm and may vary across speaking situations. To further complicate matters, individuals may speak more than one language or use multiple dialects. Little is known about the effect of code switching on voice production and identification. The purpose of this presentation is to present some of the relevant literature on voice recognition and factors related to misidentification. The role of intraspeaker variability will be discussed with a special emphasis on bilingualism and bidialectalism. Implications for voice production in augmentative and alternative communication devices will be described.

Mismatch between training and testing data is a major error source for both automatic speech recognition (ASR) and automatic speaker identification (ASI). In this paper, we first present a statistical weighting concept to exploit the unequal sensitivity of mel-frequency cepstral coefficients (MFCC) components to against the mismatch, such as ambient noise, recording equipment, transmission channels, and inter-speaker variations. We further design a new Kullback-Leibler (KL) distance based weighting algorithm according to the proposed weighting concept to real-world problems in which the label information is often not provided. We examine our algorithm in ASR with mismatch by different speakers and also in ASI with mismatch by channel noises. Experimental results demonstrate the effectiveness and robustness of our proposed method.

This paper proposes the utilization of SVD digital audio watermarking to increase the security of automatic speaker identification (ASI) systems and presents a study for the effect of watermarking on the ASI system performance. The SVD audio watermarking algorithm can be implemented on audio signals in time domain or in another appropriate transform domain and can be applied to the audio signal as a whole or on a segment-by-segment basis. The speaker recognition system works by generating a database of speaker's features using the MFCCs and polynomial shape coefficients extracted from each speaker after they are lexicographically ordered into 1-D signals. A matching process is performed for any new speaker to determine if he is belonging to the database or not, using a trained neural network. Experimental results show that the SVD audio watermarking doesn't degrade the ASI system performance severely. So, it can be used with ASI to increase security. Also, it was shown the segment by segment watermarking in the time domain achieves the highest detectability of the watermark. So, we can say that it is recommended to use SVD segment by segment audio watermarking with ASI systems implementing features extracted from the DCT or the DWT.

Speaker recognition systems based on spectral features perform well in acoustically matched and noise-free conditions. Spectral features are unsuccessful to model information about the speaker at higher levels. Prosody which represents intonation, rhythm and stress of speech, better represents speaker characteristics at higher levels. Voice disguise is a common threat to automatic speaker identification systems. It is the action of transforming one's voice purposefully to imitate or just to conceal one's identity. This work attempts to study the effects of voice disguise on prosody based speaker verification system. For this a prosody based automatic speaker verification system is implemented using support vector machine (SVM). For the extraction of prosodic features, speech is segmented into phrase-like regions by detecting long pauses regions and further syllable-like segmentation is done at the valleys of short time energy contour. In order to segment fused syllables, vowel onset point (VOP) detection is performed in such regions. Then prosodic features are extracted and used for building SVM models. To study the effect of voice disguise, normal and masked speech were recorded simultaneously for a set of speakers in the lab environment. Prosodic features for both normal and masked speech were analyzed. Using normal speech, automatic speaker verification system is implemented and testing was done using both normal and masked speech. Through this study, it is identified that prosodic features derived from pitch and duration are almost robust against voice masking but energy features are slightly affected.