Aqua Voice

Aqua Voice is a system-wide AI dictation tool for macOS, Windows and iOS. On desktop, it inserts dictated text into ordinary text fields across email, Slack, documents, browsers, terminals and coding tools; on iOS it works through an Aqua keyboard so the same account can dictate into other apps.¹² The product does more than literal speech-to-text: it removes filler, fixes grammar, structures paragraphs and adapts formatting to the destination app.¹³

Aqua's current speech stack is built around its Avalon model. The product emphasizes developer and AI vocabulary—model names, CLI commands, frameworks and code-related terms—and offers a custom dictionary for names and specialized terminology.¹⁴ Aqua also supports custom instructions so users can define writing preferences such as punctuation, tone or formatting for different workflows.¹

Deep Context is the most distinctive and privacy-sensitive feature. Aqua can use text visible on the user's screen as context so that ambiguous names, code identifiers and jargon are transcribed in the form relevant to the current application.¹ This can materially improve technical dictation, but it means users should distinguish ordinary audio transcription from the optional broader context surface.

Realtime Mode displays and refines text while the user is speaking, aimed at longer-form composition and editing.¹ The current product also exposes an Avalon batch transcription API compatible with the standard OpenAI SDK request pattern; pricing is $0.39 per hour of audio, billed per second with a ten-second minimum.⁴

Independent hands-on evidence is unusually substantive for this category. 9to5Mac's Ben Lovejoy used Aqua as his standard writing workflow for months and reported 352,000 dictated words over 105 days at an average 126 WPM. In an earlier same-passage comparison using built-in Mac microphones, Apple Dictation produced 17 errors while Aqua produced one.⁵ That is one reviewer and one comparison, not a universal benchmark, but it is stronger evidence than a brief demo.

The current Starter plan provides 1,000 free words with no credit card. Pro is $10 month-to-month or $8/month billed annually ($96/year) and includes unlimited words, custom instructions and an expanded dictionary. Max is $30 month-to-month or $24/month annually and adds Realtime Mode, the “Send it” command and early-access features.⁶

For organizations, Team is $12/user/month billed annually for 2–9 users and adds centralized billing, team-wide settings and enforceable Privacy Mode. Business is custom-priced for 10+ users and adds SSO/SAML, SCIM, advanced reporting, Zero Data Retention, team dictionaries and volume discounts.⁶ Students currently receive a 70% discount on Pro and Max.⁶

One subscription covers desktop and iOS, according to Aqua's FAQ and 9to5Mac's current hands-on report.³⁵ The product requires internet access because transcription is cloud-based.³

The Avalon API is a separate developer surface rather than simply another seat. At $0.39/hour with no seat fee, it is relevant to teams building transcription into copilots, support tooling or other software without adopting the Aqua desktop interface.⁴

Aqua has better quality evidence than many young dictation products, but the evidence should be separated by type. The company's AISpeak benchmark is vendor-created and specifically targets technical/AI vocabulary; Aqua reports about 97.3% accuracy there.⁴ Avalon also appeared on the public OpenASR leaderboard, where Aqua reports a 6.24% average word-error rate at its October 2025 debut and the leading proprietary-model position at that time.⁴⁷ Because leaderboards change and Aqua chooses how to frame AISpeak, the strongest claims should not be reduced to one timeless “97.3% accurate” product score.

9to5Mac provides useful independent validation of the everyday experience. The reviewer describes Aqua as substantially more accurate than Apple's built-in dictation, praises automatic paragraphing and voice-based editing, and continued using the product for hundreds of thousands of words.⁵ The same review identifies the core weakness clearly: server-side transcription and optional screen context create a privacy trade-off that does not exist with fully local dictation.

Product Hunt currently shows 5.0/5 from 14 reviews. Repeated praise centers on speed, technical accuracy, ease of use, dictionary customization and replacing typing across code, Slack, email and documents; reported drawbacks include occasional missed/paste failures, cloud dependence and requests for broader platform/offline support.⁸ Fourteen reviews are encouraging but still a small sample and should not be treated as a mass-market satisfaction score.

The privacy policy is current to July 29, 2026. With Privacy Mode disabled, Aqua may securely store transcript data on its servers to the extent necessary to improve the product; with Privacy Mode enabled, transcript data is not collected, though session metadata such as timestamps, device type and performance data may still be collected.⁹ Aqua states that it does not sell personal information and does not share transcripts/audio with advertising platforms. It is SOC 2 Type II certified and links a Vanta Trust Center.⁹

This creates a reasonably clear choice: ordinary users can keep transcript history and richer convenience, privacy-sensitive individuals can enable Privacy Mode, and larger organizations can buy Business with Zero Data Retention.¹⁶ None of those options makes Aqua offline: audio still has to reach the cloud transcription service.

Aqua is strongest for people who write enough that dictation can replace typing rather than supplement it. Writers, lawyers, consultants, managers and accessibility users can justify a subscription more easily when they produce thousands of words each week. Developers are an especially strong niche because Avalon and Deep Context are deliberately tuned around technical terms, code and AI prompting.¹⁴

It also makes sense for users who move between desktop and phone. Current macOS, Windows and iOS support plus synchronized dictionary/instructions make voice input more consistent across devices.¹²

It is a weaker fit for occasional dictators who can use free OS speech input, anyone who needs offline transcription, and organizations whose policy forbids audio or contextual text from leaving the device. The free 1,000-word allowance is enough to test accuracy but too small to represent sustained heavy use.⁶