An Art of Speech Recognition: A Review

Jolad, Bhuvaneshwari; Khanai, Rajashri

doi:10.1109/icspc46172.2019.8976733

Cited by 14 publications

(3 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…There are different methods for feature extraction of speech signal such as statistical methods, spectral methods, model-based methods, transform based methods, pattern recognition methods, etc. [17]. Commonly, feature representations are extracted every 10ms over a window of 20 or 30ms for speech analysis.…”

Section: B Feature Extractions Methodsmentioning

confidence: 99%

Automatic Database Segmentation using Hybrid Spectrum -Visual Approach

Gbaily

2021

The Egyptian Journal of Language Engineering

View full text Add to dashboard Cite

Nowadays automated segmentation of speech signals has been attracted many of the researchers all-over the world. Many speech processing systems require segmentation of speech waveform into principal acoustic units. In this research, TIMIT DataBase (DB) is utilized to carry on this process and justify its operation or results. Thus, this paper presents a novel method of segmentation of speech phonemes, where the proposed strategy helps in the selection of appropriate feature extraction technique for speech segmentation. There are three main techniques of feature extraction used in our research; the first technique is the Mel Frequency Cepstral Coefficient (MFCC), the second technique is known by Best Tree Encoding (BTE), while the third is Image Normalized Encoder (INE), which is a hybrid technique between the Best Tree Image (BTI), and the Convolution Neural Network (CNN) ResNet-50. Then, data are trained using a hybrid model that consists of Hidden Markov Model (HMM), and Gaussian Mixture Model (GMM) to improve the performance of automatic speech recognition. The proposed model is tested and verified against the most widely used feature MFCC plus delta and delta-delta coefficients (39 parameters) to evaluate its performance. This approach has the potential to be used in applications such as automatic speech recognition and automatic language identification. The experimental results show that BTE technique achieved the highest success rate ( ) (92.64%) than using the INE technique. However, the INE technique gives confusion success rate for Transition (Tr) and Non-Transition (NTr) of values 97.1% and 99.1%, respectively.

show abstract

Section: B Feature Extractions Methodsmentioning

confidence: 99%

Automatic Database Segmentation using Hybrid Spectrum -Visual Approach

Gbaily

2021

The Egyptian Journal of Language Engineering

View full text Add to dashboard Cite

show abstract

“…The Naive Bayes is a supervised multiclass classifier based on the probability classification Bayesian theory [47] that has been shown to perform well in small sample speech classifications [48,49]. Decision trees are a supervised non-parametric learning method that derives a heretical set of simple decision rules that are applied to the predictors to derive a predicted class.…”

Section: Statistical Modelingmentioning

confidence: 99%

Automated Detection of Dysprosody in Patients with Parkinson’s Disease

Nylén

2023

Preprint

View full text Add to dashboard Cite

Dysprosody is a commonly described feature of speech deterioration due to Parkinson’s disease. Descriptions of the tonal movements underlying dysprosody have been attempted but has not afforded automation. The current study assessed a fully automated acoustic analysis pipeline in terms of its ability to predict human raters’ perception of dysprosody in patients with Parkinson’s disease. Read speech samples of 68 speakers with PD (45 male and 23 female) aged 65.0±9.8 years were assessed by four human clinical experts in terms of dysprosody severity. The recordings were also submitted to a speech processing pipeline in which portions with speech were identified and provided with a parametrization of intonation. Five models were trained and tuned based on 75% of utterances and their perceptual evaluations in a 10-fold cross-validation procedure and evaluated in terms of their ability to predict the perceptual assessments of recordings not included in the training. The performances of models were compared to human raters’ agreement with the majority vote on a speech sample. The results showed that human raters’ assessments of dysprosody can be approximated by the automated procedure. Variability in pitch does not adequately describe the level of dysprosody due to Parkinson’s disease.

show abstract

“…Suara memiliki keunikan tersendiri dengan mempunyai sebuah parameter nilai level dengan rentang tertentu dan nilai frekuensi yang bervariasi, adapun suara juga dapat dijangkau dan tidak dapat dijangkau oleh pendengaran manusia [3]. Berdarkan teori yang dikemukakan Furui bahwa "Frekuensi Fundamental suara manusia laki-laki dan perempuan adalah dua buah kurva terdistribusi normal dengan sedikit overlapping, yang berasal dari nilai rata-rata dan simpangan baku, dari banyak populasi berbeda" [4].…”

Section: Pendahuluanunclassified

Identifikasi Jenis Kelamin Secara Real Time Berdasarkan Suara Pada Raspberry Pi

Ardiana

Dutono

Santoso

2022

JKT

View full text Add to dashboard Cite

Suara setiap penutur memiliki karakter spesifik yang unik, dipengaruhi oleh jenis kelamin, usia, emosi, dialek, dll. Penggunaan identifikasi jenis kelamin berdasarkan suara berkembang pesat, seperti pada bidang sistem kemanan, pengenalan ucapan, kecerdasan buatan, dll. Tetapi dalam pengolahan suara terdapat kesulitan yang mana karakteristik sinyal suara berdasarkan bertambahnya usia tergolong sulit dalam penentuan akurasi, dan terdapat nilai frekuensi fundamental yang overlapping di antara laki-laki dan perempuan. Penelitian ini telah dilakukan suatu pemodelan sistem identifikasi jenis kelamin berdasarkan suara secara real-time pada perangkat Raspberry Pi. Sistem ini diimplementasikan dengan 2 metode yaitu algoritma YIN dan ekstraksi fitur ciri Mel-Frequency Cepstral Coefficient (MFCC). Hasil pengujian menunjukkan keberhasilan identifikasi pada tuning parameter skema kedua lebih baik daripada skema pertama dengan mempersempit parameter frekuensi overlapping. Sistem ini menggunakan data uji dan data latih sebanyak 50 wav suara perempuan dan 50 wav suara lakilaki. Pada data uji perempuan di closed test akurasinya dari 98% menjadi 100%, lalu pada open test bermula 92% menjadi 96%. Sedangkan pengujian data laki-laki closed test dari 92% naik menjadi 98%, dan open test semula 90% naik menjadi 94%. Hal ini menunjukkan bahwa pada data yang digunakan pada penelitian ini lebih cocok menggunakan tuning parameter skema kedua untuk meningkatkan hasil akurasi.

show abstract

An Art of Speech Recognition: A Review

Cited by 14 publications

References 46 publications

Automatic Database Segmentation using Hybrid Spectrum -Visual Approach

Automatic Database Segmentation using Hybrid Spectrum -Visual Approach

Automated Detection of Dysprosody in Patients with Parkinson’s Disease

Identifikasi Jenis Kelamin Secara Real Time Berdasarkan Suara Pada Raspberry Pi

Contact Info

Product

Resources

About