Introduction
Wholesale VoIP voice quality is an engineering problem with numeric targets, not a marketing adjective. Audio quality is determined by three controllable variables: codec choice, jitter under 30ms, and packet loss under 1 percent. This guide breaks down the policies an operator must enforce to sustain a MOS above 4.0.
Key Takeaways:
- Voice quality on wholesale VoIP comes down to three controllable variables: codec choice, jitter, and packet loss.
- G.711 is the gold standard for toll-quality audio; Opus and G.722 deliver HD voice at lower bandwidth; G.729 should only be a fallback.
- Jitter under 30ms and packet loss under 1% are the thresholds that keep MOS scores above 4.0.
- QoS prioritisation, redundant transit, and real-time monitoring are what hold these targets in production, not just on paper.
- Twiching's network is engineered around these exact thresholds, with per-route dashboards so degradation is caught before customers notice.
The Three Variables That Determine Voice Quality
Wholesale VoIP voice quality is not a marketing adjective — it's an engineering outcome with numeric targets. Three variables control it end to end:
- The codec negotiated at call setup
- The jitter introduced by the network path
- The percentage of packets lost in transit
Get all three within their target ranges and a call is indistinguishable from PSTN. Miss any one of them and the symptoms are predictable: choppy audio, robotic artefacts, or a one-way call where only one side can hear the other.
Codec Selection: G.711, Opus, and G.729
G.711 is uncompressed, toll-quality audio at 64 kbps per call. It's the codec business voice has been benchmarked against for decades, and it remains the safest default when bandwidth is not the constraint.
Opus and the older G.722 are wideband codecs that deliver HD voice — audibly clearer than a standard phone call — while using less bandwidth than G.711. Both require the far end to support the same codec. They shine on SIP trunks and softphones under your own control, less so on calls that hand off to the legacy PSTN.
G.729 compresses audio down to 8 kbps per call, which matters when transit is expensive or constrained. The trade-off is a small but audible quality hit, so treat it as a fallback for constrained links, not a default.
Choosing a Codec in Practice
- Default to G.711 unless bandwidth is genuinely constrained.
- Negotiate Opus or G.722 on legs you control end-to-end for HD voice.
- Reserve G.729 for bandwidth-constrained international legs where every kbps counts.
- Never let a codec mismatch force transcoding mid-call — it's a common, avoidable source of quality loss.
Jitter: The Silent Quality Killer
Jitter is the variation in packet arrival time. Voice packets are sent at even, predictable intervals, but network congestion, route changes, and queuing delays cause them to arrive unevenly.
A jitter buffer smooths this out by holding packets briefly before playback. But a buffer sized too small drops late packets, and one sized too large adds perceptible delay. The target for wholesale VoIP is jitter held under 30 milliseconds end-to-end, measured continuously rather than sampled.
Where Jitter Comes From
- Congested transit links without QoS prioritisation for RTP media.
- Route changes mid-call as a carrier's routing engine rebalances traffic.
- Under-provisioned Wi-Fi or last-mile connections on either endpoint.
- Shared, unmanaged internet transit instead of a dedicated or QoS-marked path.
Packet Loss and Its Impact on Calls
Packet loss above roughly 1% starts to drop words out of a conversation, and above 3% a call becomes genuinely hard to follow. Unlike a file transfer, voice can't simply retransmit a lost packet — by the time it arrives, the moment has passed.
Mitigation happens at the network layer. QoS markings prioritise RTP media over other traffic, and redundant transit providers ensure a single link failure doesn't degrade every call. Session border controllers are positioned to absorb jitter and loss before it reaches the endpoint.
Measuring and Monitoring MOS in Production
Further reading: Wholesale VoIP platform
Mean Opinion Score (MOS) is the standard 1–5 scale for perceived call quality, with 4.0+ considered toll-quality and anything below 3.5 noticeably degraded. In production, MOS is estimated algorithmically per call from the codec, jitter, and loss actually observed — not from a one-time test call.
Twiching's routing engine publishes MOS, ASR, and PDD per route in a real-time dashboard. A route drifting out of spec is visible and swappable before it shows up as a support ticket.
Network-Layer Policies That Hold Quality
- QoS markings that give RTP media strict priority over other traffic on every router in the path.
- Redundant, diverse transit so a single upstream failure doesn't touch live calls.
- Jitter buffers tuned per route, not left on a single global default.
- Real-time MOS/ASR/PDD monitoring with automatic route failover on degradation.
- Session border controllers enforcing codec policy and absorbing loss at the network edge.
Conclusion
Wholesale VoIP voice quality comes down to three engineering variables held within tight thresholds: codec selection, jitter under 30ms, and packet loss under 1%. Twiching's network is built around exactly these targets — QoS-prioritised routing, redundant transit, and real-time MOS monitoring per route — so a call over the public internet holds a sustained MOS above 4.0 and stays indistinguishable from PSTN. For teams evaluating a wholesale voice partner, the practical test is simple: ask for live route-level MOS, ASR, and PDD data, not a one-time demo call.
