Using the Standard P.862 to Compare the Quality of Low-Bitrate Vocoders

Abstract

The objective methods for assessing the speech signal quality are considered: 1) Perceptual Evaluation of Speech (PESQ, ITU-T Rec. P.862), 2) Listening Quality Objective (LQO, ITU-T Rec. P.800.1). A brief description and work schemes of the PESQ technique and formulas for converting Raw MOS estimates to MOS-LQO and back are given. Low-bitrate vocoders were chosen for testing: 1) MELPe, 2) Speex, 3) Codec2. Testing of vocoders was performed on bit speeds from 700 to 4800 bps. For testing we used audio files of articulation tables with 20 records (wav, 8000 KHz, 16 bit, mono). As a result of testing, tables and graphs for Raw MOS and MOS-LQO estimates of the selected vocoders were built. When analyzing the experiments results, the conclusion is made about the effectiveness of objective methods of speech quality assessment. MELPe was identified as a promising vocoder for further development, providing at bit rates of 1200 and 2400 bps MOS quality assessment respectively 2.9...3.2 and 3.0...3.3. Speex vocoder showed comparable with MELPe evaluation results at a higher bitrate (4800 bps). Codec2 vocoder showed lower evaluation results than MELPe.