Skip to Main content Skip to Navigation

Diagnostic et évaluation automatique de la qualité vocale à partir d'indicateurs hybride

Abstract : With increasing development of new technologies (RTC, RNIS, GSM, VoIP), tele-communication services are becoming more and more diversified. To this end, telecommunication operators need to supervise in real-time the speech quality of the services they offer. Speech quality is usually evaluated from subjective experiments.. Nevertheless, such experiments are time consuming and do not allow any supervisory control. So, accurate objective models are useful to estimate the speech quality.This thesis proposes a non-intrusive model for diagnosing and evaluating speech quality using information available at the measurement point: the DESQHI model (Diagnostic and Evaluation of Speech Quality using Hybrid Indicators). It differs from existing models in terms in two main characteristics. The first one concerns the structure of the model. It is shown that speech quality can be represented as a multidimensional phenomenon incorporating three perceptual dimensions related to noisiness, speech codec and continuity. This multidimensional structure allows for a diagnostic of speech quality based on identifying the principal features affecting speech qual-ity. The second characteristic concerns the nature of indicators (signal-based and parametric) used to represent the three perceptual dimensions. Signal-based indicators use numeric information to represent the characteristics of the signal, for example, the loudness of the speech signal. Parametric indicators are obtained from the network statistics, for example, the percentage of packet loss, which gives information about the level of the discontinuity in the speech signal. This work proposes hybrid indicators (using both signal-based and parametric metrics). It is shown that they are better speech quality predictors than existing models, either parametric only (e.g. ITU-T Recommendation G.107, also known as the E-model) or signal-based only (e.g. ITU-T Recommendation P.563 model).
Document type :
Complete list of metadatas

Cited literature [109 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Friday, March 16, 2012 - 11:03:28 AM
Last modification on : Wednesday, July 8, 2020 - 12:42:26 PM
Long-term archiving on: : Monday, June 18, 2012 - 4:35:13 PM


Version validated by the jury (STAR)


  • HAL Id : tel-00679705, version 1


Adrien Leman. Diagnostic et évaluation automatique de la qualité vocale à partir d'indicateurs hybride. Autre. INSA de Lyon, 2011. Français. ⟨NNT : 2011ISAL0053⟩. ⟨tel-00679705⟩



Record views


Files downloads