[BUG]

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

🐛 Bug Report: Poor speech-to-text accuracy for Brazilian Portuguese (voice input)

📋 Description

The voice dictation feature (microphone button in the desktop app) produces frequent and severe transcription errors when used with Brazilian Portuguese speech. Words and names come out garbled, forcing the user to manually re-read and fix entire messages after dictating.

🔁 Steps to Reproduce

1. Open Claude Code desktop app
2. Set language to Brazilian Portuguese (pt-BR) speech
3. Click the microphone button and dictate a normal sentence
4. Compare the transcribed text to what was actually said

❌ Examples of Misrecognition

| What was said | What got transcribed |
|---|---|
| "a linha" (the line) | "Nicolelinha" |
| "Claude" (the assistant's name) | "Saulo" |
| "via webcam/notebook" | "webcando notebook" |

Short connector words are also frequently dropped or duplicated.

✅ Expected Behavior

Portuguese dictation should have accuracy comparable to English dictation in the same app.

⚠️ Impact

Users who depend on voice input in Portuguese (dyslexia, motor difficulty, or just preference) get a degraded experience, and end up working around it by dictating into a different app (e.g. ChatGPT) and pasting the result here — which defeats the purpose of having voice input built in.

💡 Suggestion

Please review/improve the pt-BR speech recognition pipeline, or allow selecting an alternative dictation engine.

🖥️ Environment

• OS: Windows 11 Pro
• App: Claude Code Desktop
• Language: Brazilian Portuguese (pt-BR)

What Should Happen?

Portuguese (pt-BR) dictation should have accuracy comparable to English
dictation in the same app — correctly recognizing common words, names,
and short connectors instead of producing garbled, unrelated text.

Error Messages/Logs

Steps to Reproduce

1. Open the Claude Code desktop app.
2. Make sure the system/app input language is set to Brazilian Portuguese (pt-BR).
3. Click the microphone button to start voice dictation.
4. Speak a simple, clear sentence in Brazilian Portuguese, for example:

"Claude, me explica a linha do código."

1. Stop the recording and look at the text that was transcribed into the chat input box.
2. Compare the transcribed text to what was actually said.

Expected result: the transcription should match the spoken sentence.

Actual result: the transcription is garbled and unrelated to what was said.
Examples captured during real use:

• Said: "a linha" (the line) → Transcribed: "Nicolelinha"
• Said: "Claude" (assistant's name) → Transcribed: "Saulo"
• Said: "via webcam/notebook" → Transcribed: "webcando notebook"

This is not tied to any specific project or code file — it reproduces with
any plain Portuguese sentence dictated through the microphone button, in
any conversation.

Claude Model

Sonnet (default)

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

App: Claude for Windows Version: 1.49585.0 (41ad1d) Build date: 2026-09-08 OS: Windows 11 Pro

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_