Voice Generalization: Which solution provides the best noise reduction quality during a call

Recently, Russians have started talking more on the phone — the quality of voice communication now worries users no less than the speed of the mobile Internet. Operators are paying more and more attention to the standards of speech transmission during calls — in particular, MTS was recently the first in Russia to launch a noise reduction service. The option to filter out unnecessary noise is also available in smartphones from the world's leading manufacturers. Izvestia understood what makes the voice more legible — the services of operators or the internal algorithms of phones.

How noise reduction was tested

Nowadays, even those who have almost forgotten how to dial an ordinary phone number from a notebook have begun to communicate by voice. There are more and more people talking on the subway, at public transport stops and in other crowded and noisy places. But it's not always comfortable to talk there — sometimes fellow travelers, for example, on a tram, speak in such a way that they seem to be trying to reach the interlocutor directly, without a phone.

Responding to users' request for voice communication without interference, MTS was the first and so far the only Russian telecom company to launch a noise reduction service on the market. At the hardware level, this feature is available to owners of advanced smartphones — anyone can test the carrier service. J'son & Partners Consulting conducted a comparative test of the operator's service and the options offered by gadget manufacturers. We tested the iPhone 17 Pro Max, Samsung Galaxy S26 Ultra, Xiaomi 17T, as well as Apple AirPods Pro headphones. During the tests, four typical situations with different levels of background noise were played out: an extremely loud environment, an environment with high background noise volume, and conditions with medium and low noise levels. During the testing, two experiments were conducted: in the first, the participant pronounced the text at a normal volume, in the second, he spoke softly or even in a whisper. The grades were given on a four—point scale: from 1 - "I hardly hear the interlocutor" to 4 — "I understand perfectly without effort".

Alice in Wonderland: the first headphones with an AI agent were released in Russia

How Yandex Drops differ from the flagship Samsung Galaxy Buds 3 Pro — Izvestia test

What was revealed during the tests

As follows from the J'son & Partners report, at normal speech volume, most technologies provide a fairly high level of quality: average scores range from 3.3 to 3.9 points. In the case when the interlocutors spoke on the phone normally, not in raised or lowered tones, the MTS service showed the best results (3.9 points). The iPhone 17 Pro Max (3.8) has a slightly lower indicator. This is followed by Samsung Galaxy S25 Ultra (3.5), Xiaomi 17T (3.4) and AirPods Pro (3.3).

Silent speech is a more complex scenario for noise reduction technologies: algorithms need not only to suppress external noise, but also to preserve a weak voice signal without drowning it out along with the background. As part of the testing, the best recognition quality was also obtained when using the MTS service and the iPhone 17 Pro Max, analysts at J'son & Partners concluded. MTS also took the first place in the "Silent Speech" category, scoring 3.3 points (iPhone 17 Pro Max with 2.8 points in second place — see graph).

When talking on a very noisy background (extreme volume), speech intelligibility with the operator's noise reduction enabled was rated at 3.8 points. Under the basic usage scenario, the iPhone scored 3.6 points, and with the Speech Isolation feature enabled, only 3.5. Samsung, Xiaomi and AirPods scored 3.1, 3.0 and 2.9 points, respectively.

Apple of analysis: why demand for new iPhones outstrips supply

What is the reason for the sharp rise in popularity of these smartphones in the world?

Who has less noise

According to the results of comparative testing by J'son & Partners Consulting, MTS noise reduction service showed the highest average result in the quality of speech perception — this is the final verdict of analysts. The key advantage of MTS that distinguishes this service from competitors is the highest quality of noise reduction during the transition from normal to quiet speech, according to the authors of the report. "The top 2 technologies that showed the highest quality recognition also include the basic noise reduction of the iPhone 17 Pro Max. In comparison with this option, MTS service performance is comparable at normal speech volume, but on average it is higher in the case of a whisper. The third position in scenarios with normal speech volume is occupied by the "Sound isolation" option from the iPhone, in scenarios of quiet speech (whisper) — noise reduction in Xiaomi smartphones," says the materials of J'son & Partners.

Until August 2025, users often called through messengers, says Denis Kuskov, CEO of TelecomDaily. After the introduction of restrictions on a number of foreign messengers, subscribers began to use mobile telephony much more often — its traffic increased by 35-40%, and operators are actively working to improve the quality of voice communication, the expert said. The noise reduction system is a good additional service that is interesting for some users, he believes. The operator's service in this case is better because it is more substantive — it is not made in some abstract realities, but directly on the operator's network in real conditions, Denis Kuskov noted.

Noise reduction options are in demand and relevant now — people want to speak comfortably when they are in uncomfortable conditions, said Eldar Murtazin, a leading analyst at Mobile Research Group. Modern noise reduction systems are really good at correcting sound, he says. At the same time, what the operator service does is especially noticeable on inexpensive devices, where the built-in acoustics are sometimes terrible. In fact, the operator takes over the noise reduction function, improves speech intelligibility by himself and makes it possible to hear the interlocutor well, Eldar Murtazin noted.

— Noise reduction, in my opinion, is an absolutely necessary function, because rarely anyone always manages to talk in ideal conditions, and the most urgent need for accurate voice transmission arises in conditions that are not at all ideal, — said Sergey Vilyanov, analyst at Fintech Lab. — The advantage of the operator's service is that it is not tied to the equipment. Not all subscribers have top-end smartphones, and budget devices and still popular push-button phones do not have their own noise reduction.

The cross-platform nature of the solution significantly increases its practicality, the expert believes. The telecom operator has much more raw data to train the noise reduction model than any smartphone manufacturer, Sergey Vilyanov pointed out. Moreover, this training does not stop, and in the case of new uncharacteristic noise sources, the model adapts to them, he concluded.