E-Model Parameters Estimation for VoIP with Non-ITU Codec Speech Quality Prediction
The aim of this research is to the improve performance of the E-model, which is one of the most successful non-intrusive speech quality prediction models for voice communication over a packet based network. However, the E-model still has limitations. The
- PDF / 177,231 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 63 Downloads / 183 Views
Abstract The aim of this research is to the improve performance of the E-model, which is one of the most successful non-intrusive speech quality prediction models for voice communication over a packet based network. However, the E-model still has limitations. The calculation method of the E-model is restricted to a set of voice codecs from ITU-T. This paper proposes a method to estimate two codec-related parameters that used to calculate the E-model, which are called equipment impairment factor Ie and packet loss robustness factor Bpl of the non ITU-T codec. The process to estimate both parameters uses a curve fitting method to calculate Ie values from PESQ results under various levels of network packet loss. The set of Ie and Bpl of eight narrowband codecs (G.711, G.729, GSM, AMR, iLBC, Speex, Silk, and Opus) are presented. Statistical analysis was also performed for model validation. The results show that the E-model with our Ie and Bpl parameters achieved a good accuracy and a good correspondence with PESQ MOS among the eight codecs. Keywords VoIP
⋅
E-model
⋅
QoE
⋅
Codec
1 Introduction The fundamental theorem of communication speech quality prediction has been researched and developed over many years to keep up with the rapid growth of telecommunication technology. Nowadays, packets switch networks, such as IP has become the main communication channel to carry both delay and non-delay sensitive data. Previous separate technologies such as voice, video and data commuT. Triyason (✉) ⋅ P. Kanthamanon School of Information Technology, King Mongkut’s University of Technology Thonburi, Pracha-utid Road, Bangmod, Toongkru, Bangkok, Thailand e-mail: [email protected] P. Kanthamanon e-mail: [email protected] © Springer International Publishing Switzerland 2016 P. Meesad et al. (eds.), Recent Advances in Information and Communication Technology 2016, Advances in Intelligent Systems and Computing 463, DOI 10.1007/978-3-319-40415-8_30
309
310
T. Triyason and P. Kanthamanon
nication can now share resources over the same IP network. However, IP is a best-effort network that has not been designed to guarantee a quality of service for real time communication. The network characteristics such as packet loss, delay, and jitter play a crucial role for controlling a user’s quality of experiences. In voice communication, speech quality is one of the key factors that map to a user’s quality of experience. Therefore, an appropriate speech quality measurement method is a basis for maintaining a good quality of service. Communication speech quality measurement methods have been developed with either subjective or objective approaches. The subjective measurement is an assessment done by humans and the quality is rated by Mean Opinion Score (MOS) as defined in ITU-T Recommendation P.800 [1]. The overall procedures are to ask the testers to rate the quality of a set of speech sample in a controlled environment on a five-point scale (Bad, Poor, Fair, Good, and Excellent). The main drawbacks of the subjective method are
Data Loading...
