Deep Learning Based Pathological Voice Detection Algorithm Using Speech and Electroglottographic (EGG) Signals

A novel voice pathology detection technique through deep neural networks utilizing speech and electroglottographic signals

MMHFNet: Multi-modal and multi-layer hybrid fusion network for voice pathology detection

Multimodal Voice Pathology Detection With Voice Signal and Integrated Electroglottographic (EGG) Signal From the Saarbrücken Voice Database.

Estimation of glottal closure instants (GCIs) from an electroglottograph (EGG) signal can aid in clinical applications involving the diagnosis and treatment of speech pathologies and can also serve as a ground truth to assess algorithms that estimate GCIs from speech signals. In this regard, the current work proposes a phase-difference-based approach that considers the symmetrized, differenced EGG (DEGG) signal to be the Fourier transform of an arbitrary even-signal, to estimate GCIs from EGG signals. The DEGG signal possesses sharp negative valleys at the GCIs and since the symmetrized DEGG is assumed to be a spectrum, these valleys correspond to zeros that lie outside the unit circle. The angular locations of these zeros, and in turn the locations of GCIs, can be derived from the phase-difference spectrum, since it possesses a value of around $$2\pi $$2? at these locations, the derivation of which is elaborated in the paper. The proposed algorithm is compared with the existing time of excitation generator, the high quality time of excitation algorithm, and the singularity in EGG by multiscale analysis algorithm, in terms of the identification, miss, and false alarm rates, and the identification accuracy, on normal and pathological EGG. The proposed algorithm is observed to outperform the rest with an identification rate of 98.28% in normal EGG and 96.90% in pathological EGG.

Objective: This study investigated whether different electroglottographic (EGG) parameters may be used to estimate voice quality and loading related to it, and whether voice quality is related to self-estimation of voice problems and laryngeal status. The effect of sample type was also considered. Methods: EGG and acoustic signals of a sustained vowel [a:] (90 dB_{6 cm}) and text reading during noise exposure were recorded from 93 female kindergarten teachers. Analyses were made from the sustained vowel and vowel [a:] from a stressed word. Contact quotient (CQ) was calculated. Maximum velocity of increase in contact area inferred from derivative (MDEGG) was examined. Fundamental frequency was calculated from the EGG signal. From the acoustic signal sound pressure level (SPL) was computed and pressedness in voice quality was evaluated perceptually. Self-evaluation was made with Voice Activity and Participation Profile (VAPP). Indirect laryngoscopy was also performed. Results: Pressedness in voice correlated with EGG results only for the sustained vowel, and better with MDEGG than CQ values. VAPP and laryngeal evaluation did not correlate with the acoustic or EGG parameters or with perception. Conclusions: MDEGG is worth testing as an indicator of impact stress. Sustained vowel at a controlled SPL is more suitable for voice quality evaluation than a long stressed vowel from connected speech.

Effect of Tones on Vocal Attack Time in Cantonese Speakers

Pitch-Synchronous Analysis of Human Voice

This article investigates into recently emerging approaches that use deep\nneural networks for the estimation of glottal closure instants (GCI). We build\nupon our previous approach that used synthetic speech exclusively to create\nperfectly annotated training data and that had been shown to compare favourably\nwith other training approaches using electroglottograph (EGG) signals. Here we\nintroduce a semi-supervised training strategy that allows refining the\nestimator by means of an analysis-synthesis setup using real speech signals,\nfor which GCI ground truth does not exist. Evaluation of the analyser is\nperformed by means of comparing the GCI extracted from the glottal flow signal\ngenerated by the analyser with the GCI extracted from EGG on the CMU arctic\ndataset, where EGG signals were recorded in addition to speech. We observe that\n(1.) the artificial increase of the diversity of pulse shapes that has been\nused in our previous construction of the synthetic database is beneficial, (2.)\ntraining the GCI network in the analysis-synthesis setup allows achieving a\nvery significant improvement of the GCI analyser, (3.) additional\nregularisation strategies allow improving the final analysis network when\ntrained in the analysis-synthesis setup.\n

Accurate estimation of glottal closing instants (GCIs) and opening instants (GOIs) is important for speech processing applications that benefit from glottal-synchronous processing including pitch tracking, prosodic speech modification, speech dereverberation, synthesis and study of pathological voice. We propose the Yet Another GCI/GOI Algorithm (YAGA) to detect GCIs from speech signals by employing multiscale analysis, the group delay function, and <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N</i> -best dynamic programming. A novel GOI detector based upon the consistency of the candidates' closed quotients relative to the estimated GCIs is also presented. Particular attention is paid to the precise definition of the glottal closed phase, which we define as the analysis interval that produces minimum deviation from an all-pole model of the speech signal with closed-phase linear prediction (LP). A reference algorithm analyzing both electroglottograph (EGG) and speech signals is described for evaluation of the proposed speech-based algorithm. In addition to the development of a GCI/GOI detector, an important outcome of this work is in demonstrating that GOIs derived from the EGG signal are not necessarily well-suited to closed-phase LP analysis. Evaluation of YAGA against the APLAWD and SAM databases show that GCI identification rates of up to 99.3% can be achieved with an accuracy of 0.3 ms and GOI detection can be achieved equally reliably with an accuracy of 0.5 ms.

This paper proposes a novel method for estimation of strength of excitation (SoE) from speech signal. Using lowpass filtering to remove the effect of relatively high frequency vocal tract characteristics, we estimate epoch locations. Using these epoch locations we estimate SoE. The database used for evaluation purpose is CMU-ARCTIC database consisting of electroglottograph (EGG) signals. In addition, robustness of proposed method is evaluated against signal degraded conditions for additive white noise, babble noise, vehicular noise and high frequency noise at various SNR levels. Root mean square error (RMSE) is calculated for performance measurement with SoE derived from differenced electroglottograph (DEGG) as the ground truth. Finally, proposed algorithm is compared with SoE estimated by zero frequency resonator method. Experimental results show that proposed method gives good performance for clean speech with respect to existing technique whereas in noisy environment, proposed method gives comparable results.

Prior studies in the automatic classification of voice quality have mainly studied the use of the acoustic speech signal as input. Recently, a few studies have been carried out by jointly using both speech and neck surface accelerometer (NSA) signals as inputs, and by extracting mel-frequency cepstral coefficients (MFCCs) and glottal source features. This study examines simultaneously-recorded speech and NSA signals in the classification of voice quality (breathy, modal, and pressed) using features derived from three self-supervised pre-trained models (wav2vec2-BASE, wav2vec2-LARGE, and HuBERT) and using a support vector machine (SVM) as well as convolutional neural networks (CNNs) as classifiers. Furthermore, the effectiveness of the pre-trained models is compared in feature extraction between glottal source waveforms and raw signal waveforms for both speech and NSA inputs. Using two signal processing methods (quasi-closed phase (QCP) glottal inverse filtering and zero frequency filtering (ZFF)), glottal source waveforms are estimated from both speech and NSA signals. The study has three main goals: (1) to study whether features derived from pre-trained models improve classification accuracy compared to conventional features (spectrogram, mel-spectrogram, MFCCs, i-vector, and x-vector), (2) to investigate which of the two modalities (speech vs. NSA) is more effective as input in the classification task with pre-trained model-based features, and (3) to evaluate whether the deep learning-based CNN classifier can enhance the classification accuracy in comparison to the SVM classifier. The results revealed that the use of the NSA input showed better classification performance compared to the speech signal. Between the features, the pre-trained model-based features showed better classification accuracies, both for speech and NSA inputs compared to the conventional features. The two classifiers performed equally well for all the pre-trained model-based features for both speech and NSA signals. It was also found that the HuBERT features performed better than the wav2vec2-BASE and wav2vec2-LARGE features for both speech and NSA inputs. In particular, when compared to the conventional features, the HuBERT features showed an absolute accuracy improvement of 3%–6% for speech and NSA signals in the classification of voice quality.

Electroglottography – An Update

The detection of voice pathology is a critical field in the domain of speech and healthcare, with early and accurate diagnosis being pivotal for effective treatment. Electroglottography (EGG) has been emerged as a promising tool for understanding the functioning of the vocal folds, offering valuable insights into voice disorders. This chapter highlights the current state of research in voice pathology detection using deep networks applied to EGG signals and examines various studies and methodologies in this area, emphasizing data collection and pre-processing techniques, the design of CNN architectures, training strategies, and performance evaluation metrics. Additionally, the chapter discusses the potential for further advancements, challenges, and opportunities in the field, emphasizing the importance of standardized datasets and the integration of CNN-based voice pathology detection models into clinical practice.

Aim: The aim is to compare the vocal fold vibration seen during lip and tongue trills with that seen during phonation of the sustained vowel /ε/, in terms of the periodicity of the EGG waveform and the amplitude of the EGG signal, in professional voice users. Study design: This was a quasi-experimental study. Methods: We used electroglottography (EGG) to compare the vocal fold vibration seen during tongue and lip trills with that seen during phonation of the sustained vowel /ε/, in terms of the EGG waveform periodicity and signal amplitude, in 10 classically trained, professional singers. The participants produced the sustained vowel /ε/ and performed tongue and lip trills at the same frequency and intensity. The periodicity of the waveform and the amplitude of the signal were visually analyzed by three blinded, experienced readers. To confirm the visual analysis results, we measured the jitter and shimmer of the signal and the frequency of variation in vocal fold vibration during the trill exercises. Results: The EGG waveform was classified as periodic for the sustained vowel phonation task and as quasi-periodic for the trill exercises, the vibration pattern repeating at approximately 24 Hz. Conclusion: The vibration of the vocal folds was modified according to the supraglottic movement in trills exercises.

Influence of F0 and Sequence Length of Audio and Electroglottographic Signals on Perturbation Measures for Voice Assessment