190 likes | 320 Vues
This document provides an in-depth analysis of PESQ (Perceptual Evaluation of Speech Quality), an ITU-T standard (Rec. P.862) for evaluating speech quality in telecommunications. It outlines PESQ's methodology, which compares a clean reference speech file to a degraded one, predicting a Mean Opinion Score for listening quality. The document also discusses the assessment of recording devices, their functionality, performance under load, and factors affecting speech quality. Key insights on the benefits, limitations, and practical applications of PESQ are included, ensuring comprehensive understanding across various codecs and networks.
E N D
Telecommunications Industry Association TR41.3.12-07-05-005-L v1.0 - 20050426
PESQ and Recording DevicesKevin J CrossMalden Electronics Ltd
PESQ – ITU-T Rec. P.862 • Speech Quality Metric • Compares a clean reference speech file with a degraded speech file • Predicts a Mean Opinion Score for Listening Quality
Associated Standards • P.862.1 PESQ-LQO • P.862.2 PESQ-WB • P.862.3 Use of PESQ
How PESQ Works Summary of PESQ Process
Benefits of PESQ • A Standard • Calibrated against 250000 test votes • Five years use around the world • Works across a wide range of codecs and networks
Limitations of PESQ • Noise floor and file length • Signal timing • Voiced signal compression – low bit rate codecs
Assessment of Recording Devices • Functionality – Answer, Replay Start & Stop, Interactive DTMF or Voice Activation • Performance under Load • Speech Quality of Recorded Messages
Speech Quality Modifiers • Recording media FA1000f.rst • Codec • Line interface • Network • Line interface • Phone Amplifier and Equaliser • Speaker’s transducer • Speaker’s environment
Test Interfaces • POTS • Handset port • Digital source – TDM or VoIP
Test Stimuli • Language • Phonemic Content • Duration • Quality vvn from 7905 China2f12dbpoor.rst • Gender • Spectral Content • Narrowband and Wideband material • Intermediate Reference System Send Characteristic
Acoustic Connection - Playback • Narrowband speech played back to loudspeaker • Narrowband measurement • Wideband speech played back to loudspeaker • Wideband measurement
Acoustic Connection - Record • Wideband, flat, speech to Artificial Mouth • Use narrowband IRS or mIRS reference for PESQ on playback through narrowband network • Use wideband, flat, reference for PESQ on playback through wideband network
Test Sequence • Laboratory • Measure speech levels • Measure speech quality vs speech levels • Measure speech quality vs background noise type and level • Measure speech quality vs jitter and packet loss • vvn from 7905 China2f12db 0ms 0pct.rst • vvn from 7905 China2f12db 30ms 0pct.rst • vvn from 7905 China2f12db 30ms 1pct.rst
Test Sequence • Field • Measure speech levels • Measure speech quality – two male and two female • Assess jitter and packet loss
Results • Speech quality • Speech levels • Noise • Jitter • Packet loss • Time offsets after jitter buffer
Result Accuracy • Test Equipment vvn from 7905 Thai.rstvvn from 7905 China2.rst • Test Network Eric750if.rst • Speech Levels vvn from 7905 China2.rstvvn from 7905 China2f12db.rst • Codecs vvn from vn f alaw.rstvvn from vn f ulaw.rst • Volume Controls • Automatic Gain Controls
Summary • Optimise Test Equipment and Test Network • Optimise Speech Levels • Choose most favourable Codecs • Keep Volume Controls under lock and key