Mobile voice MOS measurement: ViSQOL vs POLQA comparison guide

Voice quality remains one of the most critical QoE indicators for mobile operators, yet it is also one of the hardest to measure objectively in the field. The industry standard is MOS (Mean Opinion Score), a 1-to-5 scale originally derived from subjective human listening tests. Today, algorithmic models can estimate MOS automatically from audio samples — but the choice of algorithm has significant implications for cost, accuracy, and deployment flexibility.

This guide compares the two leading approaches: ViSQOL (developed by Google Research) and POLQA (standardized as ITU-T P.863).

What MOS measures and why it matters

MOS quantifies the perceived quality of a voice call as experienced by the end user. A MOS of 4.0+ indicates toll-quality voice; below 3.0, users typically perceive the call as poor. For VoLTE and VoNR services, operators need to track MOS across their network to:

• Detect codec degradation (AMR-WB to AMR-NB fallback)
• Identify radio conditions causing jitter, packet loss, or delay
• Benchmark voice quality against competitors
• Validate QoS after network changes or new site deployments

Manual listening tests do not scale. Automated MOS estimation on the device itself — during live calls — is the only practical approach for field operations.

POLQA: the ITU-T standard

POLQA (Perceptual Objective Listening Quality Analysis), standardized as ITU-T P.863, is the successor to PESQ (P.862). It is the recognized industry standard for objective voice quality measurement.

Strengths

• ITU-T standardization: POLQA results are accepted by regulators and in operator benchmarking reports worldwide
• Full-reference model: compares the degraded signal against the original reference, providing high accuracy for controlled test scenarios
• Super-wideband support: handles narrowband (300-3400 Hz), wideband (50-7000 Hz), and super-wideband (50-14000 Hz) audio
• Proven track record: decades of deployment in telecom testing labs

Limitations

• Licensing cost: POLQA requires per-channel or per-device licensing from OPTICOM GmbH, typically ranging from 5,000 to 15,000 EUR per endpoint depending on volume
• Reference signal required: as a full-reference model, POLQA needs access to the original clean signal, which complicates real-world field deployment
• Closed implementation: the algorithm is proprietary, limiting customization and integration flexibility
• Hardware dependency: many POLQA deployments require dedicated test equipment (probe units) rather than running on standard smartphones

ViSQOL: the ML-based alternative

ViSQOL (Virtual Speech Quality Objective Listener) was developed by Google Research and published as an open research project. It uses machine learning models trained on large datasets of speech and audio quality assessments.

Strengths

• No per-channel royalties: ViSQOL does not carry per-device or per-channel licensing fees, making it dramatically cheaper to deploy at scale
• Dual-mode support: speech mode at 16 kHz for voice calls, and audio mode at 48 kHz for wideband audio quality assessment
• ML-based architecture: the model can be retrained or fine-tuned for specific codecs, network conditions, or use cases
• Lightweight computation: ViSQOL runs efficiently on mobile ARM64 processors, enabling on-device MOS estimation without offloading to a server
• Research-backed: published in peer-reviewed literature with transparent methodology

Limitations

• Not ITU-T standardized: ViSQOL results may not be accepted in regulatory contexts that specifically require P.863 compliance
• Full-reference model: like POLQA, ViSQOL compares degraded audio against a reference signal (though reference-free extensions exist in research)
• Less industry adoption: fewer operators have historical ViSQOL baselines compared to POLQA/PESQ

Head-to-head comparison

Criterion	ViSQOL	POLQA (P.863)
Standardization	Google Research	ITU-T P.863
Licensing model	No per-channel royalties	Per-channel/device license
Cost per endpoint	Included in SDK pricing	5,000 - 15,000 EUR
Speech bandwidth	16 kHz speech, 48 kHz audio	NB/WB/SWB
On-device mobile	Yes (ARM64 native)	Typically requires probe hardware
Accuracy vs subjective MOS	High correlation (r > 0.9)	High correlation (r > 0.95)
Customization	Retrainable ML model	Closed algorithm
Regulatory acceptance	Limited	Widely accepted

Which one should you choose?

The answer depends on your use case:

Choose POLQA if:

• You need results accepted by national regulators (e.g., ARCEP, Ofcom, BIPT)
• Your organization has budget for per-device licensing
• You operate a lab environment with controlled reference signals

Choose ViSQOL if:

• You need to scale MOS measurement across dozens or hundreds of devices
• Your budget does not allow 5-15K EUR per endpoint in licensing
• You want on-device MOS estimation running natively on Android smartphones
• You are building an internal QoE monitoring tool and do not need regulatory compliance

For most field engineering teams and operators running internal benchmarks, ViSQOL delivers comparable accuracy at a fraction of the cost.

How HiCellTek implements voice MOS

HiCellTek’s MOS Voice SDK integrates ViSQOL natively on Android ARM64 devices. During a live VoLTE or VoNR call, the SDK captures the downlink audio stream, runs ViSQOL inference on-device, and produces a per-segment MOS score in real time. No external probe hardware, no server-side processing, no per-channel royalties.

The MOS scores are tagged with the corresponding RF conditions (RSRP, SINR, serving cell) and Layer 3 context (RRC state, codec negotiated), enabling direct correlation between radio quality and voice experience.

Conclusion

Both ViSQOL and POLQA are capable voice quality measurement algorithms, but they serve different market segments. POLQA remains the gold standard for regulatory reporting, while ViSQOL offers a practical, cost-effective path to large-scale mobile voice QoE monitoring.

HiCellTek’s MOS Voice SDK brings ViSQOL-based MOS measurement to any Qualcomm Android device for 4,490 EUR/year — a fraction of a single POLQA license. Review the full SDK details and licensing on the pricing page.