Google has pulled one of those apps out of its hat that arrives almost without making a sound.But it has everything it takes to become an essential tool for many professionals. It's called Google AI Edge Eloquent, and although its name sounds a bit long, its idea is very simple: you dictate with your voice, and the AI cleans, reorganizes, and polishes the text in real time so that it's ready to send or use with hardly any tweaking.
Unlike other launches filled with announcements and conferences, Eloquent has appeared rather discreetly on the iOS App StoreHowever, behind this discretion lies a very serious undertaking: bringing advanced artificial intelligence directly to mobile devices, without relying on the cloud, with a clear focus on privacy and the quality of the final text. Let's break down exactly what Google AI Edge Eloquent is, what it's used for, and why it's generating so much buzz among users who rely heavily on voice dictation.
What is Google AI Edge Eloquent and how does it work?
Google AI Edge Eloquent is a voice dictation application with artificial intelligenceDesigned to transform your voice into highly polished text almost instantly. The key is not only that it transcribes what you say, but that it understands that people don't speak the same way they write and automatically bridges that gap.
Instead of simply making a literal transcription, The app rewrites your improvised sentences on the fly.It corrects typical speech errors and formats the text to make it look like you wrote it calmly on the computer. In other words, if you speak haltingly, hesitating, repeating words, or correcting yourself mid-sentence, the result you see on screen is a fluid and professional text, without the chaos of spoken language.
To get it, Eloquent relies on Gemma generative AI models installed directly on the deviceThese models are capable of real-time speech recognition, fluency correction, punctuation correction, and text restructuring. The goal is to save you time by eliminating the need to review each paragraph afterward.
Google describes the application as a tool designed to “closing the gap between how we actually speak and professional writing”This puts it in a different category compared to the standard dictation built into mobile phones, which normally only writes down exactly what it hears, including all the hesitations, filler words and stumbles.
AI-powered dictation on the device: Gemma and offline mode
One of the great advantages of Google AI Edge Eloquent is that it works locally on the iPhonewithout needing to send the audio to external servers. The app downloads a lightweight Gemma model (on the order of several hundred megabytes) that runs using the phone's own processor and Neural Engine.
This means that You can use the app completely offline, even in airplane mode.Once the model is installed, voice recognition and text cleaning are performed entirely on the device. There are no audio uploads to the cloud, no remote processing, and no dependence on network coverage. It is especially useful for journalists working in the field, professionals with sensitive information, or anyone who simply doesn't want their notes circulating on external servers.
Google has equipped Eloquent with a Clear switch to activate a “total offline mode”With this setting, the app ensures that absolutely all processing happens locally, without exception. In practical terms, this means you can be on a plane, in a garage with no signal, or in a room where you don't want a connection, and still dictate long texts without interruption.
The Gemma models used by Eloquent are part of a Google's family of generative AI models with open weightsThis means their parameters can be downloaded and run on various devices, including mobile phones. They are designed to be lightweight, efficient, and capable of handling tasks such as transcription, summarization, text generation, and basic reasoning without requiring a large cloud infrastructure.
The “Edge AI” philosophy that Eloquent represents marks an interesting shift in Google’s strategyThe focus has shifted from a "cloud-first" approach to placing much greater emphasis on on-device execution. In a way, the app demonstrates the extent to which Apple hardware can be pushed to its limits with proprietary models, competing head-to-head with the iPhone's native dictation features.
Cloud mode with Gemini: optional extra polish
Although Eloquent is designed to work primarily offlineThe application also offers an optional cloud-based mode. If the user activates it, part of the text processing is sent to Gemini models hosted on Google's servers, which can provide an even more refined finish to complex sentences or particularly long documents.
In this “Cloud” mode, The app continues to capture the audio on the device, but the final text processing is handled by Gemini.This can translate into better handling of complicated grammatical structures, finer detection of nuances, and even more natural writing when dictation is chaotic or full of changes of idea.
Google presents this duality as a balance between privacy and maximum text qualityBy default, everything is done locally, but if you need that extra touch of polish (for example, for formal reports, emails to important clients, or complex legal documentation), you can activate cloud processing with a simple setting in the app.
It is important to emphasize that Gemini activation is entirely voluntary.Those with strict confidentiality requirements can maintain local-only mode without sacrificing most of the tool's advantages. The idea is that it's not an "all or nothing" situation: each user chooses the extent to which they are willing to sacrifice privacy for added quality in the finished text.
What historical problem does it solve: nobody speaks like they write
For years, big tech companies have tried to convince us that voice would be the natural way to talk to our deviceswith assistants like Google Assistant, Siri, or Alexa. However, these assistants never achieved the widespread use that was expected, and one of the main reasons has been their inability to convert spontaneous speech into truly useful text.
The underlying problem is obvious: We don't speak the same way we writeWhen we dictate, we tend to mix up phrases, make odd pauses, use filler words, correct ourselves mid-sentence, and jump from one idea to another without much structure. Traditional dictation systems, even relatively good ones, simply copy all of that onto the screen, resulting in a text that is difficult to read and requires extensive proofreading.
Google AI Edge Eloquent specifically addresses this limitation: It focuses on understanding improvised speech and transforming it into professional text from minute one.It doesn't just recognize words; it interprets the context, detects where you've gotten stuck, decides which parts are unnecessary and how to reconstruct the sentence so that it makes sense in writing.
In practice, this means that You can dictate almost exactly as you think, without trying to "speak perfectly"If you add a “well, wait, maybe this is better…” or repeat a word while searching for the right expression, Eloquent can clean up that noise and leave you with a smooth, flowing text. For anyone who takes quick notes on the go, records project ideas, or drafts entire emails while walking, the improvement in ease of use is remarkable.
In this area, the app positions itself as a direct competitor of other advanced AI-based dictation and transcription solutionsas the Whisper from OpenAI or services like Wispr Flow and SuperWhisperbut with a strong focus on local execution and privacy, something that many professional users value above all else.
Key functions: cleaning, formatting, and customization
One of Eloquent's most striking features is its ability to "clean up" spontaneous speech.The application automatically detects and removes filler words such as "uh," "mmm," "uhm," and other hesitating expressions that slip in without us noticing when we think out loud.
Besides filler words, The app corrects punctuation and agreement errors. These are typical of speech but not of writing. For example, it rearranges the sentence when you correct yourself mid-sentence, adds commas and periods where appropriate, and adjusts verb tenses if you've accidentally changed them. All of this happens in real time as you speak.
Another strong point is that The tool offers different output formats for the textYou are not forced to stick with a flat block of words, but can choose from several options depending on how you want to use the dictation:
- Summary of key points to extract the essentials from a meeting or a long idea.
- Formal text for professional emails, reports, or corporate documents.
- Short version If you need something concise and direct.
- Extended version to further develop the ideas and obtain a longer text.
Personalization goes a step further with the option to create a custom and editable dictionaryEloquent allows you to manually add proper nouns, technical jargon, acronyms, or industry-specific terms (e.g., medical, legal, or engineering) so that the model recognizes them correctly and does not try to "correct" them to something similar.
In addition, the app can Import vocabulary from your Gmail accountBy analyzing your email history, Eloquent learns the names of clients, projects, products, or common phrases you frequently use. This way, when you dictate something with a complicated last name or the unusual name of an internal project, it's much more likely to spell it correctly the first time.
With use, The AI adapts to the way you speakIt improves with your pronunciation patterns, your usual grammar, and your style preferences. This continuous learning capability means that, as you use it, the transcripts become increasingly accurate and tailored to your way of speaking.
Additional tools: metrics and session management
Google AI Edge Eloquent goes beyond transcription and formattingIt also incorporates features designed for those who frequently teach and want to monitor their productivity or easily review previous sessions.
On one hand, the app It records the total number of words and the dictation speed. (words per minute) of your sessions. This can be useful for journalists, writers, or professionals who want to measure how much text "output" they achieve in a given time or check if they have improved their work pace.
You can also check your dictation history and search through previous recordings to locate specific segments. If you use the app to take meeting notes, jot down quick ideas, or draft documents, this feature makes retrieving information much easier without having to manually search through each transcript.
In combination with the format options (summary, formal text, etc.), These tools make Eloquent more than just a simple dictation tool.It's close to being a voice writing assistant that organizes, measures, and puts order to the vast amount of content you can generate by speaking.
Privacy and sensitive professional sectors
Privacy is one of the fundamental pillars of Google AI Edge Eloquent's approachUnlike many services that rely almost entirely on cloud servers, this app is designed specifically for users who are most reluctant to have their voice data leave the device.
In local mode, Audio recordings and transcripts are not sent to GoogleEverything is processed using the Gemma model installed on the iPhone itself, providing an extra level of peace of mind for sectors where confidentiality is critical: law firms, clinics and hospitals, consultancies, companies with trade secrets, etc.
Google allows, optionally, grant the app permission to access certain Google Workspace data (for example, Gmail emails) with the sole purpose of enriching the contextual dictionary and improving the understanding of specific names and terms. This authorization is configurable and can be revoked at any time.
The company insists that The app is free and has no daily usage limits.You don't need a Google Workspace subscription to take advantage of basic offline dictation features. Integrations with more advanced Gemini models may vary depending on the region and Google's policies at any given time, but the core local dictation functionality remains free.
Regarding the comparison with other solutions, Eloquent further reduces the "privacy gap" compared to what Apple offers natively.Although Apple has long boasted about its on-device processing capabilities, Google's offering literally allows you to put your phone in airplane mode, cut off all network connections, and still have fully functional AI dictation—something highly valued by journalists, executives, and professionals who handle particularly sensitive information.
Availability, languages, and device requirements
Currently, Google AI Edge Eloquent is only available for iOS. It is available through the Apple App Store and is offered primarily in English. Currently, the app does not officially function in Spain or other regions of the European Economic Area due to regulatory approval requirements.
Google has indicated that Arrival in the UK, EEA and Switzerland is restricted for regulatory reasonsHowever, they are actively working to obtain the necessary approvals to expand into these areas later. It's reasonable to expect that, once these procedures are completed, the app will gradually reach more markets.
As for the hardware, Eloquent requires an iPhone with a powerful Neural Engine and a recent version of iOSThe application is optimized for modern models such as iPhone 16 Pro, iPhone 17 and later, with iOS 26.0 or higher, since the intensive local processing of Gemma models demands good computing power to offer low latency.
For now, There is no final version for Android or other operating systems.Although previous App Store descriptions mentioned Android support, which was later removed, it seems Google is using Apple's closed hardware ecosystem as an ideal testing ground for its Edge AI SDK before expanding to other platforms.
In addition to Eloquent, Google has another app on the App Store called AI Edge GalleryThis app serves as a testing environment for the Gemma model family, including variants such as Gemma 3n and Gemma 4. It allows you to run the models entirely on your device, test chat and step-by-step reasoning functions, and compare the performance of different model sizes on your hardware.
Gemma and Google's family of open models
The technology that powers Google AI Edge Eloquent is based on Gemma, a family of generative AI models that Google has developed with the same technological base as Gemini, but designed so that developers and companies can run it in their own environments, without always depending on Google's infrastructure.
Gemma models are considered to be “open pesos”This means that its parameters can be downloaded, adjusted, and deployed on proprietary hardware under specific usage conditions. This allows it to be integrated into local applications, enterprise servers, and, as in the case of Eloquent, directly into smartphones.
In practice, Gemma is useful for tasks such as summarizing texts, answering questions, reasoning, generating content, and processing transcripts.The lighter variants are designed to run on devices with limited resources, such as mobile phones or laptops, while the larger models (26B and 31B parameters, for example) are intended for desktop computers and servers.
Gemma 4, one of the newest families, includes models specifically adapted to phones (like E2B and E4B) and leverages very wide context windows, up to 128K tokens, expandable to 256K in some large variants. This means they can handle very long documents or extended conversation sessions without losing track.
In the case of Eloquent, Google would be using compact versions of Gemma, possibly quantifiedThese optimizations achieve a balance between text quality, battery consumption, and response speed. Thanks to these improvements, it's possible to enjoy a highly advanced AI-powered dictation experience without needing a constant cloud connection.
The existence of AI Edge Gallery as an accessible lab on iOS reinforces the idea that Google wants to turn Gemma into the foundation of a local AI ecosystemwhere applications like Eloquent are just the tip of the iceberg for end users, while developers experiment with even more complex variants in the background.
Google AI Edge Eloquent thus positions itself at a very interesting point in the current landscape of AI applied to dictationIt combines extremely fast on-device processing, deep speech cleaning, multiple output formats, and remarkable respect for privacy. For anyone who drafts emails, reports, meeting notes, or entire ideas by speaking into their mobile phone, it represents a leap in convenience that goes beyond simple "voice recognition" and comes close to having a true ghostwriter installed in their pocket.
