Feature request: real-time streaming text-to-speech as responses are generated

Preflight Checklist

• [x] I have searched existing requests — closest is #45863 (closed as not planned), but that asked for read-aloud of completed messages. This request is specifically for streaming speech.
• [x] This is a single feature request

Problem Statement

I use voice/dictation to interact with Claude Code hands-free. Today the only option is OS-level read-aloud (e.g. macOS Spoken Content), which speaks a message only after it has fully completed. During long agentic turns this means long silences followed by a wall of speech — and it often ends up reading stale/previous messages. There is no way to hear Claude 'thinking out loud' in real time as it types and runs tools.

Proposed Solution

Stream text-to-speech incrementally as the assistant's text output is generated (sentence-by-sentence or chunk-by-chunk), so the user hears the response in real time, including brief status updates between tool calls. Ideally:

• A toggle (e.g. /voice or a setting) to enable spoken output
• Speak assistant text as it streams, not after turn completion
• Skip or summarize tool output; speak only assistant prose
• Available in CLI and desktop app

Alternatives

• OS read-aloud (laggy, reads whole transcript, no streaming)
• Claude mobile app voice mode (not available for Claude Code workflows)

Additional Context

This would make long-running agentic sessions usable hands-free — the user can follow progress by ear while doing other things.