MOS Score Calculator

How this estimate works

This calculator uses a simplified version of the narrowband E-model (ITU-T G.107) to estimate conversational voice quality from four inputs: one-way network delay, jitter buffer size, codec, and packet loss.

The model combines these inputs into an R-factor — a scalar that represents the combined effect of all quality impairments — then converts that R-factor to an estimated MOS-CQE using a standard polynomial mapping. Both values are shown in the result.

Total delay (Ta). The model first computes the total one-way mouth-to-ear delay: Ta = network delay + jitter buffer + codec overhead. The codec overhead in this simplified implementation is bundled into the codec preset (20 ms for G.711, 35 ms for G.729A). For example: 40 ms network delay + 40 ms jitter buffer + 20 ms G.711 overhead = Ta of 100 ms.

Results can be inaccurate if the network delay value you enter already includes codec framing or playout-buffer delay — that would cause the calculator to double-count those contributions to Ta.

Delay impairment (Idd). When Ta is 100 ms or less, the pure delay impairment is zero. Above 100 ms, it increases nonlinearly using the G.107 formula. Delay becomes a significant impairment factor as Ta approaches and exceeds 150 ms — the threshold ITU-T G.114 identifies for noticeable quality degradation.

Equipment and packet-loss impairment (Ie,eff). The effective equipment impairment combines the codec's baseline compression impairment (Ie) with the effect of packet loss, weighted by the codec's packet-loss robustness factor (Bpl). Higher Bpl means the codec degrades more slowly as packet loss increases.

R-factor. R = 93.2 − Idd − Ie,eff. The 93.2 baseline reflects this simplified implementation's fixed narrowband defaults. The full E-model includes additional parameters (room noise, sidetone, echo impairment) that this calculator does not expose.

What MOS-CQE means

The E-model outputs a predicted conversational quality score on the standard 1-to-5 MOS scale. Because it is computed from parameters rather than from audio signals or human listeners, the output is technically an estimated MOS-CQE.

These interpretive ranges are general descriptions for planning purposes only. They are not authoritative pass/fail thresholds. Actual perceived quality depends on factors this model does not capture.

MOS-CQE vs measured MOS

There are three distinct types of MOS in use in the VoIP industry:

MOS-LQS (Listening Quality Subjective)

Gathered by human listeners rating speech in controlled laboratory conditions, as described in ITU-T P.800. The original definition of MOS. Requires real audio samples and trained raters.

MOS-LQO (Listening Quality Objective)

Produced by PESQ (ITU-T P.862) or POLQA (ITU-T P.863). Compares an original audio signal against a degraded version algorithmically. Requires actual audio streams.

MOS-CQE (Conversational Quality Estimated)

Estimated from network and codec parameters using the E-model. Does not require audio. Suitable for planning and monitoring. This is what this calculator produces.

Calling an E-model output a "measured MOS" conflates these distinct methods. This calculator produces MOS-CQE only.

What affects the estimate

The four inputs affect the result in the following ways:

• Higher network delay increases Idd (pure delay impairment), lowering R and MOS-CQE. The effect is zero up to a Ta of 100 ms and grows nonlinearly beyond that.
• Larger jitter buffer adds to Ta and therefore increases Idd the same way as network delay. This reflects the trade-off inherent in jitter buffering: larger buffers reduce packet loss from late arrivals but add delay impairment.
• Codec choice determines Ie (baseline compression impairment) and Bpl (packet-loss robustness). G.711 (PLC) starts with Ie = 0 — no baseline codec impairment — but has a higher Bpl than G.729A, meaning it handles random packet loss more robustly in this model. G.729A starts with Ie = 11, reflecting its higher compression ratio.
• Higher packet loss increases Ie,eff and therefore lowers R and MOS-CQE. The rate of degradation depends on the codec's Bpl value.

Methodology and assumptions

This is a simplified planning estimator, not a complete implementation of ITU-T G.107. The key implementation decisions are:

• Narrowband model only. Based on ITU-T G.107 (narrowband). Wideband codecs (G.722, Opus) require ITU-T G.107.1 and are not modeled here.
• Delay impairment. Uses the Idd component of G.107 (eq. 7-27) with default delay-sensitivity class. Talker-echo and listener-echo impairment components are not exposed; this simplified estimator does not model the full set of echo-related E-model parameters.
• Baseline R. Fixed at 93.2, reflecting simplified narrowband defaults for the parameters this calculator does not expose (room noise, sidetone, quantization noise). The full E-model uses additional parameters.
• Packet-loss model. BurstR = 1 (random, non-bursty loss). Bursty packet loss may degrade quality more than this model predicts.
• Codec presets. Ie and Bpl values are implementation presets based on standards-referencing technical sources (G.113 Appendix I planning values). G.711 presets assume packet-loss concealment is active. G.113 Appendix I primary text was not directly accessed during research.
• Codec overhead. Codec framing/algorithmic delay is bundled into each codec preset (G.711: 20 ms; G.729A: 35 ms) and added to Ta automatically.

Limitations

This calculator has the following known limitations. Understand them before using results for network decisions:

• Simplified planning estimator — not live audio measurement or standardized testing
• Narrowband model only (G.107) — does not support wideband or fullband codecs
• Does not perform P.800 subjective listening tests, PESQ, or POLQA analysis
• Does not model talker echo or listener echo (Idte/Idle components)
• Assumes random packet loss (BurstR = 1); bursty loss degrades quality more than this model predicts
• Assumes packet-loss concealment is active on G.711 endpoints
• May overstate delay impairment if the entered network delay already includes codec or playout-buffer contributions
• Does not account for room noise, sidetone, or line noise — fixed narrowband defaults are used
• Codec overhead (packetization + algorithmic delay) is a bundled preset; actual overhead depends on endpoint configuration