[BUG] German Read Aloud ignores voice selection on macOS and web; works on iPhone

### Preflight Checklist

- [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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?

German Read Aloud always uses the female “Sanft” voice, even when the male “Hell” voice is selected.

The “Hell” preview correctly plays the male voice. The selection also persists when reopening settings. However, clicking the play button beneath a German response always plays the female voice, regardless of the selected style.

This happens in the Claude macOS app across Chat, Code, and Cowork, and in the web interface on laptops. It also happens in new conversations and on every laptop I have tested.

On iPhone, selecting “Hell” correctly produces the male voice during Read Aloud. English voice selection also works.

German is explicitly selected in the voice settings. This report concerns the play button that reads an existing response aloud.

### What Should Happen?

German Read Aloud should use the selected voice style and match the settings preview:

* “Hell” should play the male voice.
* “Sanft” should play the female voice.

The behavior on macOS and web should match the working behavior on iPhone. If a voice is unsupported on a particular platform, the settings should clearly indicate this instead of accepting a selection that has no effect.

### Error Messages/Logs

```shell

```

### Steps to Reproduce

1. Open the Claude macOS app or the Claude web interface.
2. Open the voice settings and explicitly select German.
3. Select the “Hell” voice style.
4. Play its preview and confirm that the male voice is audible.
5. Start a new conversation and ask: “Schreibe drei kurze Sätze über das Wetter.”
6. Click the play button beneath the German response.
7. Observe that the female “Sanft” voice plays instead of the selected male voice.
8. Reopen settings and confirm that “Hell” is still selected.
9. Switch between “Hell” and “Sanft” and repeat playback. Both selections produce the same female voice.

For comparison, repeat the German Read Aloud test on iPhone: the selected male voice works correctly there.

### Claude Model

None

### Is this a regression?

Yes, this worked in a previous version

### Last Working Version

Unknown. The previous voice setup handled the spoken language automatically. I first noticed this problem after the transition to language-specific voice settings, and German voice selection has not worked correctly for me since then.

### Claude Code Version

Claude Desktop 1.46388.4 (macOS app; CLI version not collected)

### Platform

Anthropic API

### Operating System

macOS

### Terminal/Shell

Terminal.app (macOS)

### Additional Information

Environment:

* Claude Desktop version: 1.46388.4
* macOS version: 26.6.2
* Subscription: Claude Max, using the official Claude app and website
* Voice language: German
* Selected voice style: Hell
* Voice actually played: Sanft
* Interface: Desktop app GUI and web; no terminal involved in reproduction

Comparison:

* macOS Chat, Code, and Cowork: affected
* Web interface on laptops: affected
* New conversations: affected
* German “Hell” preview: correct male voice
* German Read Aloud on iPhone: correct selected voice
* English voice selection: works

I have already contacted support but have not received a reply.

Potentially related reports:

* https://github.com/anthropics/claude-code/issues/84979
* https://github.com/anthropics/claude-code/issues/85239

These describe related voice problems, but my specific symptom is that German Read Aloud consistently ignores the selected style on macOS and web while the preview and iPhone playback work correctly.

Could you confirm whether this is a known regression and whether a workaround is available?

No source files, tests, or entry points are named. Start by reproducing the German Read Aloud behavior on macOS and web, comparing the settings preview and iPhone playback, then trace how the selected voice style reaches response playback. Done means “Hell” and “Sanft” consistently produce their selected voices, or unsupported platforms clearly reject the selection.

Written by the indexing model from the issue text.