feat(voice): implement speech-friendly response formatter for voice mode output

[ayush31010]

What would you like to be added?

A formatForSpeech() function in packages/core/src/voice/responseFormatter.ts that converts markdown/ANSI-formatted tool output and LLM responses into speech-clean plain text for TTS playback in voice mode.

Required transformations:

• Strip all markdown syntax (bold, italic, code fences, blockquotes, headings, links, lists)
• Strip ANSI escape codes
• Abbreviate deep absolute paths (/a/b/c/d/e/file.ts:142 → …/e/file.ts line 142)
• Summarise large JSON blobs as (JSON object with N keys)
• Collapse stack traces to first line + (and N more frames)
• Truncate outputs above ~500 chars with … (N chars total) suffix

Proposed API:

export function formatForSpeech(
  text: string,
  options?: { maxLength?: number; pathDepth?: number },
): string;

Why is this needed?

The voice mode architecture skeleton (PR #20779) introduced VoiceModeController and the core interfaces. Before any text can be streamed to TTS, it must be converted to speech-clean plain text.

Every existing tool result and LLM response is formatted for visual rendering. A response like:

Error: ENOENT: no such file or directory

at /home/user/project/packages/core/src/tools/file-utils.ts:142:7

would be read aloud as:
"asterisk asterisk Error asterisk asterisk backtick E N O E N T colon no such file or directory backtick greater-than at slash home slash user slash..."

Without this formatter, VoiceModeService cannot produce natural-sounding output. It is a hard prerequisite for the Live API streaming integration.

Additional context

• Architecture skeleton PR: feat(cli): introduce experimental voice mode architecture skeleton #20779
• RFC/Architecture discussion: [RFC] Architecture Proposal: Hands-Free Multimodal Voice Mode (GSoC 2026) #20456
• GSoC 2026 Project 11: Hands-Free Multimodal Voice Mode

Files to create:

• packages/core/src/voice/responseFormatter.ts
• packages/core/src/voice/responseFormatter.test.ts

[ayush31010]

/assign

[gemini-cli]

👋 @Solventerritory, thanks for your interest in this issue! We're reserving self-assignment for issues that have been marked with the help wanted label. Feel free to check out our list of issues that need attention.

[ayush31010]

@gsquared94 @bdmorgan

Implemented in PR #20989.

The formatForSpeech() function is now available at packages/core/src/voice/responseFormatter.ts with full test coverage (36 tests).

What was implemented:

• All 7 transformations described in the issue (ANSI stripping, fenced code blocks, stack trace collapse, markdown syntax, path abbreviation, whitespace normalisation, truncation)
• jsonThreshold added as a third option alongside maxLength and pathDepth — controls when a JSON blob gets summarised vs kept as-is
• A follow-up fix was pushed to correct a bug where Windows paths on non-C: drives (e.g. D:\, E:\) were being reconstructed with the wrong drive letter

Verified:
npx vitest run packages/core/src/voice/responseFormatter.test.ts
✓ 36 tests passed

Zero new runtime dependencies — pure TypeScript, Node stdlib only.

[ved015]

/assign

[gemini-cli]

👋 @ved015, you've been assigned to this issue! Thank you for taking the time to contribute. Make sure to check out our contributing guidelines.

[ayush31010]

@jacob314 Can I get assigned for this issue since I raised this issue and also have an active pr on it

[ved015]

hi @Solventerritory no issue i will unassign from this issue you can work on it but please follow the open source guidelines only raise PR for issue on which you are assigned and which are marked help wanted dont raise PR unless that issue is marked as help wanted by the maintainer😁
Still thank you for your cooperation i will unassign now

[ved015]

/unassign