[BUG] Voice dictation duplicates interim transcript segments (Chinese) in Claude Code Desktop

Environment

• Claude Desktop app (Microsoft Store / MSIX) 1.22209.0.0, Windows 11 Pro (build 26200)
• Bundled Claude Code: 2.1.209 (AppData\Roaming\Claude\claude-code\2.1.209\claude.exe)
• Dictation: built-in mic button in the desktop app input bar, "hold to record" mode
• "language": "chinese" set in ~/.claude/settings.json (bug occurs both with and without this setting)
• Microphone: USB Audio Device (also reproduces with other input devices selected)

Describe the bug

When dictating in Mandarin Chinese, the inserted transcript duplicates interim segments: a prefix of the sentence is inserted first, then the sentence is inserted again including that same prefix. On longer utterances nearly every phrase appears 2x (sometimes 4-5x), making the transcript unusable.

Examples (spoken vs. transcribed)

1. Spoken: 现在语音功能是不是恢复了

Transcribed: 现在语音功能是不是现在语音功能是不是恢复了

1. Spoken: 这样我需要提交一个工单。你帮我提交一下呗。

Transcribed: 这样我需要提交一个工单这样我需要提交一个工单。 你帮我提交一下呗你帮我提交一下呗。

1. Longer utterances (~30s) repeat almost every phrase, e.g.:

我用的是这个自带的我用的是这个自带的。 语音识别功能语音识别功能但是它 一直有这种问题你可以看我这句话一直有这种问题你可以看我这句话也是 同样的...

The pattern is always A + AB — an interim partial result (A) is committed to the input, then the finalized segment (AB) is appended instead of replacing the interim text.

Expected behavior

The transcript should be inserted once, with interim partial results replaced by the finalized text rather than duplicated.

Notes

• Reproduces on essentially every utterance longer than a few words; short phrases (<~3s) are usually fine.
• Speech is Mandarin Chinese; app UI language is Korean.
• Word-level recognition is otherwise mostly correct — the problem is the duplication, not the recognition itself.