2.png){kind=small}
Battle of the DNNs: Who Did AI Hearing Better?
Updated: 3 days ago
Signia MaX 7, Oticon Reveal, and the science behind the sales pitch
Two contenders. Six neural networks. One dinner table that refuses to take turns.
Oticon Reveal arrives with two cooperating AI systems. Signia MaX arrives with four specialist DNNs and a headline announcing 54 times more processing power. Somewhere, a hearing-aid wearer just wants to hear the punchline before everyone starts laughing.
So who did DNN better? We dug through technical papers, human studies, and independent laboratory results. There are winners here. They just need the correct trophies.
Round 1: four versus two. Somebody bring a calculator.
A deep neural network, or DNN, learns patterns from training examples. Inside a hearing aid, that can help estimate which parts of a messy mixture belong to speech, noise, or other sounds. The fitted device then uses those estimates to guide processing. Your hearing aid has not acquired opinions about the restaurant.
Signia's Acoustic Intelligence has four disclosed DNN tasks: detecting speech, detecting your own voice, analyzing noise, and analyzing the sound scene. Oticon Reveal uses Speech AI to identify speech and Context AI to help balance speech and surrounding sound. [3, 10, 11]
Signia wins specialist headcount and dedicated own-voice specialization. That is meaningful architecture, but four networks do not automatically outperform two. Network size, training, accuracy, power constraints, and the processing they control matter.
Also, Signia's four AI-branded processing features are not another four DNNs. Oticon's 4D sensors are not four secret networks either. Marketing has given the number four quite enough work already. [3, 12, 56]
Round 2: the 54-times-powerful microphone drop
The actual Signia claim is 54 times more processing power than IX, alongside twice the memory and roughly 50% greater energy efficiency. That comparison is with Signia's predecessor, not Reveal. It does not mean 54 times better hearing, and the public release does not give a reproducible chip benchmark. [1]
Think of a larger restaurant kitchen. More cooking capacity can support more dishes at once; it cannot promise your favorite meal arrives 54 times sooner.
MaX's Acoustic Intelligence analyzes sound 100 times per second, or every 10 milliseconds. Reveal describes Speech AI and Context AI updates every 2 milliseconds. Those are different tasks and schedules, not matched measurements of the total delay from microphone to ear. [2, 10]
Reveal wins the shorter published analysis interval. MaX has a substantial advertised compute upgrade over IX. Neither figure settles which wearer understands more words.
More compute plausibly enables richer simultaneous analysis and spatial processing. Whether that becomes smoother speaker changes or cleaner speech is a listening-test question. Horsepower is exciting. We still need to drive the car.
Round 3: the training-data weigh-in
Signia reports 40 million sound samples, speech in 11 languages, and six music genres across Acoustic Intelligence. That is useful training-breadth information, not 40 million examples for each network. Nor does it mean MaX translates your French dinner guest. [2, 5]
Oticon describes individually training Reveal's networks with realistic sound scenes captured using a 32-microphone spherical array. The often-repeated 12 million sound scenes belongs to the earlier More generation; it is not a comparable current Reveal training total. [11, 18]
So a 40-million-versus-12-million scoreboard would compare different generations and different counting units. Longer scenes, shorter windows, and augmented copies are not interchangeable training currency.
No training-quality champion. A useful contest would test unfamiliar voices, accents, reverberation, and noise on the same held-out recordings. The public disclosures do not supply that benchmark.
Round 4: new technology, family resemblance
MaX is a real development of IX. AX established separate processing for focused sound and surroundings. IX added dynamic conversation streams. MaX adds learned coordination, three additional narrower focus streams, a Multi-Stream Denoiser, and a Contrast Enhancer. Conversation support did not suddenly appear because somebody printed “AI” on the box. [3, 6, 7]
Oticon has a lineage too: More introduced its DNN processing; Real improved comfort features; Intent added DNN 2.0 and 4D sensing. Reveal adds learned speech detection alongside a revised context network. Context AI is described as a retrained successor to Intent's processing, now using improved input from Speech AI. [18, 19, 20, 28]
Both earn an award for a meaningful upgrade. MaX expands specialist analysis and conversation processing. Reveal adds speech/context cooperation. Calling either an entirely unrelated invention misses the foundations; calling either an AI sticker misses the changes.
Round 5: the loudest person at dinner already wins enough
In a noisy group, detecting speech is only the first problem. Which voice should stand out, from which direction, and for how long?
MaX uses its learned analysis to guide expanded conversation streams. The denoiser adds binaural noise reduction; the contrast component permits more linear gain during speech to improve separation from the background. Those are interacting processing changes, not a magic consequence of the number four. [3]
Reveal's Voice Clarifier uses speech information to guide directional reductions while protecting target speech. Its operating region can be Front, Wide, or Full. Within that region, the technical description prioritizes the loudest of simultaneous speakers; unreliable speech detection can trigger a frontal fallback. [10]
Useful engineering? Yes. Mind reading? No. “Speech” is an acoustic category. “The person I actually came here with” involves intention.
Both systems also aim to preserve useful surroundings. Oticon makes context central to its design; Signia retains a surrounding stream and varies noise relief with the scene. Neither brand gets to claim the other simply deletes the world. [10, 57]
No universal dinner-table champion. Speaker geometry, movement, fitting, and the scene all matter. We would like the hearing aid to require evidence before taking the loudest person's side.
Round 6: your own voice would like a word
“I sound strange” can describe two different problems: physical occlusion from the ear coupling, or the way the aid amplifies and processes the wearer's voice.
MaX's own-voice DNN uses spatial near-field information to recognize the wearer. OVP AI then adjusts processing during speech. It removes the personal in-clinic own-voice calibration needed by earlier Signia OVP generations. [3]
Signia wins this disclosed specialization and simpler calibration workflow. That is a concrete advantage for fitting and first impressions.
But a better voice detector cannot remove every physical effect of sealing an ear. We still need a controlled Reveal comparison with comparable couplings to crown the more comfortable own-voice experience.
A fifth imaginary DNN called “Please Stop Sounding Like a Bucket” would still have to negotiate with the ear tip.
Round 7: the results get a referee
Signia earns a real human-study win
An Oldenburg study tested 30 experienced hearing-aid users. MaX 7 achieved a mean SRT80 of -5.2 dB, versus -4.2 dB for 7IX, -3.7 dB for unnamed competitor A, and -2.4 dB for unnamed B. SRT80 is the speech-to-noise ratio needed to repeat 80% of test words; lower is better. Twenty-nine of 30 performed best with MaX. [4]
MaX wins that reported contest. The manufacturer-associated study used proprietary first fits, closed couplings, manual speech-in-noise programs, and forward-facing heads. The competitors were unnamed. We cannot secretly replace “Brand A” with “Reveal.”
A separate 26-person Signia-associated study reported 6-21 percentage-point word-score advantages and up to 1.8 points less effort on a 10-point scale. Its SNR was individualized using MaX's speech threshold, so those conditions matter. [5]
Oticon's impressive numbers need the right units
Reveal's speech/context brief reports up to 2.3 dB more output-SNR improvement than Intent, using manikin recordings and modeling rather than human word testing. Different setups prevent subtracting that figure from Signia's acoustic results to manufacture a brand advantage. [14]
The “60% stronger brain response” headline comes from a cortical-response study: 17 participants completed both Reveal and Intent conditions, listening passively to synthetic speech-like stimuli. The approximately 60% increase concerned a mid-frequency F-ratio, a statistic for detecting the neural response against EEG variability. [13]
Interesting physiology, yes. It does not mean 60% more words, 60% less effort, better cognition, or a win over MaX. A bigger EEG statistic is not a firmware update for the wearer.
The independent lab awards Reveal a different trophy
HearAdvisor's available tuned-fitting results put Reveal 1 ahead of the MaX configuration it tested: 3.34 versus 2.76 for speech in noise, 4.78 versus 3.64 in quiet, 5.00 versus 4.60 for feedback handling, and 3.50 versus 2.90 for music. [30, 31]
Reveal wins those published independent metrics. These are complete-device laboratory scores using acoustic measurements and models, not a human crossover trial of isolated DNNs. The listed MaX technology tier is unclear; MaX used a vented Sleeve and Reveal a Bass Dome. Both tuned fits targeted NAL-NL2 for the standardized N3 hearing loss. [30, 31, 32, 33]
MaX leads the reported own-voice/occlusion score, 3.70 versus 1.90. However, that metric measures passive occlusion for these products, with the aid powered off. It does not test OVP AI. The ear coupling gets that trophy; the neural network cannot collect it. [34]
The judges' decision
Signia wins: more disclosed specialist DNN tasks, dedicated own-voice detection with simpler calibration, and the reported human speech contests against IX and unnamed competitors.
Reveal wins: the shorter published analysis interval and the available independent laboratory speech and feedback scores in the tested configurations.
Still unresolved: whose DNNs generalize better, which preserves context better across brands, and which current premium aid wins a controlled MaX 7-versus-Reveal 1 human trial.
That is a useful answer to “who did it better?” Both brands have serious engineering. Particular evidence favors particular winners. No public result here crowns an intrinsically superior neural network.
For the specialist, the next valuable evidence is a matched crossover: comparable fitting targets, verified real-ear output, controlled couplings, moving talkers, overlapping speech, effort ratings, and transparent analysis. For the wearer, compare the fitted systems in the conversations you actually need to follow.
Network count cannot rescue a poor fitting. Channel count is not model size. And feedback control, spatial processing, and the ear tip are working just as hard as the glamorous AI.
The DNNs can argue about their résumés. Your ears get to judge the conversation.
Want the engineering details? The companion deep dive retains the full methods, predecessor comparisons, training discussion, chip-claim analysis, and 58-source research trail.
A few fair questions
Do four DNNs make Signia MaX 7 better than Oticon Reveal?
Four specialist networks and two cooperating networks describe different designs. Network count alone cannot establish better hearing. The disclosed studies do not provide a controlled, matched human trial that isolates and ranks these two DNN systems.
Which hearing aid leads the independent speech results?
Reveal 1 leads the available HearAdvisor tuned-fitting speech scores for the tested configurations. The MaX tier is unclear and the ear couplings differ, so this is a laboratory-result win, not a universal premium-device clinical verdict.
Can I compare hearing aids locally with Clarity?
Clarity offers free hearing tests and hearing aid guidance in Osprey, Englewood and St. Petersburg. Ask about a suitable fitting and current demonstration availability before your visit.
Your table. Your ears. Your local fitting.
Bring us the conversations you miss, not just the AI headline. Clarity Hearing Care can help you compare fitted hearing aids and decide which technology suits your hearing and daily life. Ask which models are available to demonstrate.
Osprey: Visit our Osprey office at 2101 S. Tamiami Trail, Osprey, FL 34229, convenient for Nokomis, Casey Key and Palmer Ranch. Call 941-477-2101.
Englewood: Visit our Englewood office at 2828 S. McCall Rd, Suite 43, Englewood, FL 34224, convenient for Wellen Park, Rotonda West and Placida. Call 941-475-9909.
St. Petersburg: Visit our St. Petersburg office at 6701 38th Ave. N., Suite A, St. Petersburg, FL 33710, convenient for Tyrone, Gulfport and Pinellas Park. Call 727-624-6701.
Request your free hearing test. Let your ears cast the deciding vote.
Research references
Optional reference reading. Research cutoff: September 29, 2026. Numbering follows the full 58-source technical companion.
[5] Korhonen et al. (Aug 2026). Enhancing group conversations in noise with RTCE AI. ORCA white paper.
Find your Clarity office
See all three locations together, then choose an office to explore its latest 3D neighborhood view.
Written by Brad Wagner, HAS, BC-HIS
Senior Hearing Aid Specialist and Co-Owner, Clarity Hearing Care
Comments