Speech quality examples based on ITU-T Recommendation P.800

The reference speech files that are used in the construction of the degraded examples consist of two short sentences seperated by a silence of about 1 second, total duration is about 7 seconds.
They have been recorded at close distance in an anechoic room and the noise in the silent parts has been suppressed.
These reference files are in line with the references as used in the development of P.862 PESQ and P.863 POLQA .

With each degraded example the estimated Mean Opinion Score (MOS) is given as would have been obtained in an absolute category rating P.800 (section B.4.5) listening test,
using high quality headphone reproduction with the following five point quality scale:
5 = excellent
4 = good
3 = fair
2 = poor
1 = bad


Reference speech quality examples, bandwidth 14 kHz:
no degradation 20-14000 Hz: Igbo female   Dutch male, MOS is about 4.8


Degraded speech quality examples, linear filtering:
marginal degradation 50-5000 Hz:   Igbo female   Dutch male, MOS is about 4.0
      If you don't hear the difference with the reference you have:
        1) a bad end-to-end play back link (from server to ear including the DA conversion, amplifier and loudspeaker/heaphone)
        2) bad ears, either due to aging (>60 years?) or noise induced hearing loss
low degradation         300-3400 Hz: Igbo female   Dutch male, MOS is about 3.0
medium degradation 300-2000 Hz: Igbo female   Dutch male, MOS is about 2.0


Degraded speech quality examples, packet loss:
low loss:         Igbo female   Dutch male, MOS is about 3.0
      Note that this degradation is less audible with loudspeaker reproduction due to time domain smearing
medium loss: Igbo female   Dutch male, MOS is about 2.0
severe loss:     Igbo female   Dutch male, MOS is about 1.0


Degraded speech quality examples, room reverberation:
low reverberation:         Igbo female   Dutch male, MOS is about 3.0
medium reverberation: Igbo female   Dutch male, MOS is about 2.0


Degraded speech quality examples, pink noise:
very low noise: Igbo female   Dutch male, MOS is about 4.0
low noise:         Igbo female   Dutch male, MOS is about 3.0
medium noise: Igbo female   Dutch male, MOS is about 2.0


Degraded speech quality examples, amplitude clipping:
marginal clipping: Igbo female   Dutch male, MOS is about 4.0
low clipping:         Igbo female   Dutch male, MOS is about 3.0
medium clipping: Igbo female   Dutch male, MOS is about 2.0
severe clipping:     Igbo female   Dutch male, MOS is about 1.0