Launch HN: Aqua Voice (YC W24) – Voice-driven text editor

Hacker News Discussion ↗

A new voice-driven text editor, Aqua Voice, aims to go beyond traditional dictation by understanding user intent and directly editing documents, which many see as a breakthrough for accessibility, productivity, and people who think better aloud. Early users praise the demo’s ability to handle corrections and formatting by voice, but raise concerns about latency, reliability at scale, accent and language support, and the lack of native apps or deep integrations with editors and operating systems. There is active debate over pricing models, privacy and data retention, and whether the technology should eventually be offered as an API or OS-level utility to support use cases from coding and note-taking to medical documentation and education.

Overall reaction

  • Many commenters found the demo “wow”-level impressive and immediately useful, especially compared to past dictation tools.
  • Several subscribed on the spot or said they’d “happily pay” if it integrated better with their workflows.
  • Others were excited but bounced due to latency, clunkiness, or missing features, canceling quickly.

Use cases and target users

  • Strong interest from:
    • People with RSI, neuropathy, disabilities, and dyslexia who rely on or prefer voice.
    • Students and knowledge workers for papers, email, and note‑taking.
    • Healthcare (doctors, dentists, radiologists) where dictation is already ubiquitous.
    • Developers and power users wanting IDE/editor integration (VS Code, JetBrains, Obsidian, Joplin, Notion, Raycast).
  • Aspirational use cases: background “day-long” note-taking, walking/cycling monologues, meeting/whiteboard transcription, interview review, recipes, speeches, screenplays.

Interaction model: natural language vs commands

  • Product leans toward “command-less” intent understanding from speech alone.
  • Some users see this as the right long‑term direction; others argue natural language alone will never be as efficient as explicit commands.
  • Repeated requests for:
    • A hybrid model: natural language + user‑defined commands/macros/aliases.
    • Better support for structured formats (tables, screenplay, code) and domain vocabularies (acronyms, custom dictionaries).

Technical / product details

  • Uses a custom “fusion” model for intent and fine‑tuned GPT‑4 for rewriting/transformations.
  • Browser app today; Mac app in flux; strong demand for:
    • Native desktop and mobile apps.
    • System-level “keyboard” style integration across all text fields.
    • API access so others can build plugins and native wrappers.
  • Issues reported: Firefox audio errors, Edge problems, mic selection friction, accent errors (e.g., Scottish), AirPods quality, and sometimes over‑conservative editing.

Pricing, tokens, and business model

  • Confusion and dislike of “tokens” as a user-facing unit; suggestions to use words, minutes, or time‑based trials instead.
  • Some want pay‑as‑you‑go/credits instead of subscriptions.
  • Concern that the free allocation feels too small to meaningfully evaluate daily use.

Privacy, data, and platform concerns

  • Multiple users ask what audio/text is sent to the cloud, how long it’s retained, and whether it’s used for training.
  • Some would pay extra for strict deletion/no‑training modes.
  • Strong pushback on Google OAuth‑only signup; requests for email-based accounts and non‑Google options.
  • Desire for local/offline models for privacy, cost, and reliability; some suggest “personal use local, business use paid cloud.”