Assessment of objective voice quality over best-effort networks

VoIP calls transferred over dedicated bandwidth or QoS capable networks is a cost-effective alternative for PSTN in large enterprises. However, the calls made over the best effort network, such as the global Internet, suffer packet loss and jitter. In some VoIP-codecs, such as ITU G.723.1 and G.729a, there are built-in recovery mechanisms for concealing packet-based errors in the audio/speech stream. These recovery mechanisms can conceal up to 5% packet losses without significant quality degradation, as shown in this article. The 5% quality degradation is approximately within 0.5 MOS scale when compared to the original signal. Beyond 5%, the speech quality will drop gradually. The overall quality of MOS scale 3 can be maintained even with 14-17% packet loss rates. The influence of delay variation or jitter cannot be eliminated with the concealment algorithms unless the jitter time exceeds packet loss indication delay. The influence of jitter is not critical below 20 ms but beyond 20 ms limit its influence will decrease the speech quality very steeply. This suggests that packet losses can be recovered in normal conditions, but the influence of jitter must be eliminated somehow. The interleaving and piggybacking-based stream manipulation enhances speech quality in packet dropout situations. The guaranteed delay over the whole Internet would enhance the possibilities of VoIP to achieve success.