← Echo Blog

Voice Keyboard vs Voice App: Why the Extension Wins

By Xiang · April 16, 2026 · 4 min read

Most voice-to-text tools are standalone apps. You open them, speak, then copy-paste the result into whatever app you actually needed.

Echo takes a different approach: it's a keyboard extension. No copy-paste. No app-switching. Just tap the mic anywhere you'd type.

Here's why this architectural choice matters more than it sounds.

The Hidden Cost of App-Switching

Let's trace a typical voice-to-text workflow in a standalone app like Otter or Wispr:

1. Open Messages → want to reply with voice
2. Switch to Otter/Wispr
3. Tap record, speak
4. Wait for transcript
5. Tap to copy
6. Switch back to Messages
7. Tap paste
8. Send

Eight steps. App-switch twice. Copy-paste once.

Now the same workflow in Echo:

1. Open Messages
2. Tap mic on Echo keyboard
3. Speak
4. Text appears in reply field
5. Send

Five steps. Zero context-switch. Zero copy-paste.

Why Does This Add Up?

Each app-switch costs you 1-2 seconds plus a mental transition. Copy-paste adds another 1-2 seconds of "did it copy the right thing?" friction. Over a day of voice-typing messages, emails, notes, this burden adds up fast.

More importantly: the workflow friction is why most people don't voice-type even when they have a voice app installed. The activation energy is too high. A keyboard extension removes that friction entirely.

The Other Hidden Win: Typo Correction

Voice ASR isn't perfect. Even the best models (Seed-ASR 2.0, Whisper Large) will occasionally:

• Miss homophones (that vs thats)
• Mangle uncommon names
• Drop words in noisy environments

With a standalone voice app, fixing a typo is painful — you have to copy-paste back into the app, edit, copy out again. Most people just accept the error.

With a keyboard extension, the correction happens in place. You see the bad word in your Messages reply, tap it with the built-in keyboard, fix it, done. Voice and text editing live in the same interface.

Why Other Apps Don't Do This

Building a keyboard extension on iOS is harder than building a standalone app:

• Sandboxing: Keyboard extensions can't access the microphone directly. You need IPC with a host app.
• Memory limits: Extensions have stricter memory caps. No loading a 500MB Whisper model.
• Full Access: Users have to manually toggle "Allow Full Access" in system settings, which Apple warns about.
• Testing: Extensions are harder to debug — every keystroke matters across every app.

The easier path is a standalone voice app. That's why Otter, Wispr, and most others went that way. Building a keyboard was a 10x more complex engineering problem — but it's the only way to make voice input actually fit into real workflows.

The Architecture

For the curious: Echo is structured as:

• EchoApp — Main iOS app (handles settings, recording, ASR pipeline, AI polish)
• EchoKeyboard — Keyboard extension (accepts typing + triggers voice via deep link)
• App Group — Shared UserDefaults between them for IPC (settings, results)
• Deep Link — echo://voice — keyboard hands control to main app for recording

When you tap the mic:

1. Keyboard writes intent to App Group
2. Deep link opens main app
3. Main app starts recording (mic access OK here)
4. ASR transcribes, AI polishes
5. Result written to App Group
6. You swipe back to Messages
7. Keyboard reads result, inserts text

Complicated to build. Invisible to use. That's the whole point.

Try It

Download Echo (iOS) →

Questions or feedback? Reach me on X @EchoVoiceApp.

— Xiang, solo maker of Echo