voize

Proprietary domain specific speech recognition and an on device large language model, both built in house by founders with systems engineering backgrounds. The technical choice to run inference locally rather than calling a hosted model is itself evidence that model work is the company rather than a wrapper around someone else's API.

This remains the most consequential gap in the record, and the second pass sharpens what is at stake without closing it.

Nothing was located describing the review gate between spoken input and a committed entry. That matters more here than for an office visit scribe because of what this product writes into. Medication administration records and vital signs are legally significant clinical documents where an error is a patient safety event rather than a wording problem. Fluid intake feeds clinical decisions in a frail population. Emergency documentation is an action rather than a record.

The second pass adds a dimension no physician office scribe faces. The vendor states that the product recognises residents and assigns documentation to the right person, presenting that as a feature. In a facility with many residents, correct attribution is not a convenience but the difference between a vital sign landing on the right chart and the wrong one. A misassigned medication entry is a safety event affecting two people at once, the resident whose record gains an entry that did not happen and the resident whose record loses one that did.

So there are two gates to establish rather than one. What a caregiver confirms about the content before it is committed, and what they confirm about the resident it is attributed to. Both should be answered per documentation category, since a narrative care report and a medication administration entry do not warrant the same treatment.

The speech input compounds it. The system is described as handling colloquial and imperfect German and converting it into formalised documentation, which is a real capability and also means the committed entry may differ substantially in wording from what was said.

More architecturally specific than most in this category. voize states what the model is, an on device large language model, where it runs, locally on smartphones, and what it was built to handle, namely medical shorthand, regional accents and multilingual speech patterns. That is a description a buyer can test rather than a set of adjectives.

Held at B because the headline accuracy claim of 99.2 percent is published without any methodology, denominator, reference standard or breakdown, and an accuracy figure that precise invites the question of what exactly was measured.

The architecture answers the central question structurally rather than by disclosure, which is the strongest form of answer available on this axis. The model is described as an on device large language model running locally on a smartphone and functioning without a constant internet connection, so speech is processed where it is spoken and no third party model provider sits in the inference path at all.

A buyer can verify that claim by testing the application offline, which is more than can be said for any policy commitment in this lane. Where a chain does not exist, there is nothing to enumerate and nothing to leak. Held below the top grade for two reasons, one of which a buyer should read carefully.

The model is described by type and location but not by identity, so no base model, provider or version is named, and an on device model is very often a third party open weight model adapted for the task rather than one built from nothing.

The second is the boundary of the local processing claim itself: inference is local, but the resulting structured documentation still synchronises into the care record system, so the protection covers the audio and the processing rather than the finished record and whatever handles it downstream. Ask which base model the on device system derives from, and for the chain behind the synchronisation path rather than the capture path.

Above the vendor testimonial norm of this category because an academic institution was involved. A study conducted with the Charite in Berlin, one of Europe's largest university hospitals, reported roughly 30 percent time savings on documentation, and a separate reported figure puts the saving at an average of 39 minutes per shift. Deployment scale is genuine rather than aspirational at more than 75,000 nurses across over 1,100 facilities.

Held at B rather than A because no peer reviewed publication was located, the study design and cohort were not retrieved, and this index does not assert details it has not seen. A published Charite paper would move this to A.

The strongest privacy architecture located in this category, because it is structural rather than contractual. The model runs on the caregiver's device and functions without a constant internet connection, so speech is processed where it is spoken rather than transmitted to a vendor for inference. That removes an entire class of exposure rather than promising to manage it, and unlike a retention policy it is verifiable by the buyer simply by testing the app offline.

Read the boundary precisely: inference is local, but the resulting structured documentation still synchronises into the care record system, so this protects the audio and the processing rather than the finished record.

The earlier assessment treated this axis as not applicable, on the basis that deployment was Germany and Austria under the European data protection regime, while noting a stated intention to enter the United States market. That intention has been executed. The vendor now describes a standard United States deployment and markets to skilled nursing and long term care facilities there, so the axis applies.

What a counterparty can verify is thin in the usual way. No business associate agreement template, no scope statement and no subprocessor list were located.

The architecture materially changes the shape of the relationship though, and that deserves stating rather than treating this as an ordinary gap. The vendor states that in its United States deployment speech recognition and documentation generation run on the caregiver's device and only structured documentation is transmitted into the customer's record system. A vendor designed so that protected health information does not come to rest in its own environment holds far less than a cloud processing competitor, which reduces both the breach surface and the scope of what an agreement needs to govern.

It does not remove the requirement. Creating and transmitting protected health information on a covered entity's behalf establishes the relationship regardless of where processing happens, and the application, the synchronisation path and any support access remain the vendor's. A buyer should establish what the agreement says about each of those rather than assuming the on device design makes it unnecessary.

One note on the home market position, which does not transfer. The vendor's European compliance covers the general data protection regulation and the two German church data protection acts governing the Catholic and Protestant bodies that operate much of German social care. That is a precise and appropriate posture for that market and says nothing about the United States rule.

No named or dated attestation, no SOC 2 report of either type, no ISO 27001 and no trust centre were located in a second pass.

What is published instead is a list of data protection regimes the product is stated to comply with, and it is more specific than the category norm. Alongside the European data protection regulation, the vendor names the two German church data protection acts that govern the Catholic and Protestant bodies operating a substantial share of German social care. Naming those rather than stopping at the general regulation shows an accurate understanding of who its customers actually are. They remain statutes rather than attestations, and compliance with a law is a self assessment unless someone independent has examined it.

The architecture changes what an attestation would need to cover, and that is worth stating rather than treating the absence as identical to a peer's. Where speech recognition and documentation generation run on the caregiver's own device and only structured documentation reaches the customer's record system, the vendor holds materially less than a cloud processing competitor does. An examination of this vendor's estate would therefore be examining a smaller surface. It would still be worth having, because the mobile application, the synchronisation path, the model distribution mechanism and whatever backend supports them all remain the vendor's responsibility.

The scale of the customer base and the expansion into a second market raise the expectation rather than lowering it.

Ask which report is held, of which type and period, and specifically whether its scope covers the on device components and the synchronisation path rather than only the backend.

No clearance claimed and none required, and no device pathway attaches in either market served.

The flag raised in the earlier assessment stands and the second pass makes it more pressing rather than less, because the United States expansion has happened and the documentation scope is wider than a narrative note.

The product generates entries across medication logs, vital signs, fluid intake, wound documentation with photographs, care reports and emergency documentation. Several of those sit further from narrative documentation than an office visit note does. A medication administration record is a legally significant document in its own right, vital signs feed clinical judgement directly, and emergency documentation accompanies an action rather than describing a completed encounter. None of that makes the product a regulated device today. It does mean the classification question deserves asking rather than assuming, particularly as scope widens, and a buyer should not carry the ambient scribe answer across to a product that writes structured clinical values.

What actually governs in the home market is long term care documentation requirement and inspection rather than device regulation, where documentation quality is assessed as part of the care quality regime and failures surface as findings against the facility rather than as device incidents. The facility remains accountable for its records whatever produced them.

For the United States market the equivalent frame is the conditions of participation and survey process applying to skilled nursing facilities, which likewise place the obligation on the operator. A buyer should establish how the product supports evidencing that obligation, since the value of accurate contemporaneous documentation in that setting is partly defensive.

Treats the category's central failure mode as a design requirement rather than a disclaimer, which is rare and worth crediting. voize states its model was built to understand regional accents and multilingual speech patterns alongside medical shorthand.

That matters unusually much for this specific user population: care work in Germany and Austria is disproportionately performed by multilingual and migrant staff, so a speech model that degrades on non native accents would fail hardest on exactly the workforce it is sold to support. Held at B because no performance breakdown by accent, language or speaker group has been published, so the design intent is stated rather than demonstrated. Note the pattern across this index: the two vendors engineering for accent variation rather than disclaiming it, voize and Corti, are both European and both build their own models.

The grade rests mainly on the regime rather than on the vendor's own commitments, and the distinction is worth stating because it applies to the whole European cohort in this index. Operating under European data protection law, the person recorded holds enforceable statutory rights over their own personal data, including rectification of inaccurate records, with a supervisory authority behind them and obligations that reach processors and not only the customer.

That is a genuine correction route for the affected person, and it exists whether or not the vendor chooses to offer one. The vendor side contribution is the architecture rather than a promise. Local processing on the caregiver's device removes an entire class of exposure structurally, and a buyer can test it. Within the European cohort this sits at the lower end, and the reason is the published accuracy figure.

A stated 99.2 percent arrives with no methodology, denominator, reference standard or breakdown, and a number given to one decimal place invites precisely the question of what was measured. An unfalsifiable figure is not a commitment however precise it looks, because there is no way for a customer to establish it was not met. No warranty, indemnity or remediation obligation was located. Ask for the accuracy definition and sample, and for what the vendor commits to when documentation is wrong.

Integrates into existing care documentation systems through APIs and interfaces with data synchronisation of documentation and client information, with Vivendi PD named as a supported system. Note the segment difference: these are long term care documentation platforms rather than hospital electronic health records, so integration breadth is measured against a different and largely German speaking market than the rest of this category. Coverage outside that market is not yet evidenced.

The clearest answer to the where does the data go question available anywhere in this index, because the answer is that it does not go. Inference runs on the device in the caregiver's hand and continues working without a constant internet connection, which also solves a practical problem specific to the setting: care homes and hospital wards have poor connectivity in exactly the rooms where documentation happens.

Compare the two other strong answers in this category, Corti offering sovereign cloud or on premise deployment and Lyrebird confining everything to one national jurisdiction. voize is the only one that removes the transmission of speech entirely.

No published rate card located. Sold to care facility and hospital operators rather than to individual clinicians, so pricing is negotiated at the organisation level. One market signal rather than a price: some care providers now cite voize in job listings as a recruitment advantage, which suggests operators are valuing it partly as a retention instrument rather than purely as a documentation cost line.

Covers a workforce this index otherwise barely reaches. Where nearly every competitor documents a physician office visit, voize covers the nursing and caregiving day across care homes and hospitals: documentation, shift handovers, wound documentation with imaging, vital signs, movement logs, medication administration, scheduling and emergency alerts. Shift handover in particular is a high risk communication event with no equivalent in outpatient documentation.

The only comparable nursing capability in this index is the Dragon Copilot nursing edition, which is a module of a physician product rather than a product designed for the role. Geographic reach is currently Germany and Austria.