feat(desktop): add pitch-preserving TTS playback speed

[johnmatthewtennant]

Context

Generated speech currently plays at a fixed 1x rate. Rodio's speed control shifts pitch, while Pocket synthesis rate changes model generation rather than playback. This keeps playback speed independent from Pocket-only voice settings and preserves the existing per-chunk streaming path.

Summary

Adds a locally persisted global 0.75x to 1.50x generated-speech playback setting, with a 1x reset and pitch-preserving Signalsmith time stretching. The setting layers into Voice settings while remaining separate from Pocket voice selection and synthesis.

Changes

• Adds a 1x bypass and pitch-preserving Signalsmith processor for each complete Pocket playback chunk before rodio append.
• Preserves the fixed 20 ms device lead-in outside time stretching, keeps one contiguous append per chunk, and retains the existing post-DSP barge-in and voice-change cancellation check.
• Keeps Pocket time-to-first-playback streaming: the first synthesized chunk plays without waiting for the full response.
• Adds global local persistence with a conservative 0.75x to 1.50x range, serialized native writes, cross-platform atomic replacement, initialization-race protection, intent coalescing, and Reset to 1x.
• Places the control in Voice settings independently of Pocket enablement and voice selection, including preview playback and the desktop E2E bridge.
• Adds DSP pitch, order, onset, tail, and latency tests, production lead-in and constructor integration tests, UI persistence-race tests, and an end-to-end processing benchmark.
• Adds ssstretch 0.1.0 (Signalsmith Stretch), MIT licensed. It is bundled at build time; normal use performs no download.

Related issue

N/A. No related issue was supplied.

Testing

• Normal pre-push hooks on the integrated head passed: branch-skew, Desktop static guards, 3,736 Desktop tests, 824 Mobile tests with 1 skipped, all root Rust unit suites, and full Desktop Tauri tests.
• Integrated-head Desktop Tauri result: 1,833 unit tests and 3 diagnostics passed, 14 ignored, 0 failed.
• Focused Rust playback tests: 8 passed; Settings voice transition and selection: 8 passed; Settings constructor: 1 passed.
• just desktop-tauri-clippy: passed with warnings denied.
• just desktop-test: 3,733 passed on the reviewed pre-restack head. The integrated pre-push run passed 3,736.
• just desktop-check: passed, including file-size, px-text, and pubkey-truncation guards.
• just desktop-typecheck: passed.
• cargo deny check licenses: passed. ssstretch and Signalsmith are MIT licensed; warnings are existing unmatched allowances for MITNFA, OpenSSL, Unicode-DFS-2016, and bzip2-1.0.6.
• Fresh E2E build plus pnpm exec playwright test tests/e2e/voice-settings.spec.ts --project=smoke: 2 of 2 passed in 3.8 seconds.
• Full Builderbot, Kalvin, and minimize-diff review completed. Selective post-base Builderbot and minimize reviews found zero findings, and the prior final Kalvin rerun was clean. No findings remain.
• Remaining manual validation: launch the real Tauri app; open Settings, then Voice; listen to Mary and Marius preview and active-huddle speech at 0.75x, 1.25x, and 1.50x; verify pitch, first phoneme, and tail; interrupt speech to verify immediate cancellation; restart to verify persistence; use Reset and confirm 1.00x. This has not been performed on the headless machine.

Reviewer-reproducible examples

Run the exact production-helper benchmark:

cargo run --release --manifest-path desktop/src-tauri/Cargo.toml --example tts_speed_probe

Observed output on the final DSP implementation:

0.75x: processing=20.63ms for 10s (0.206% realtime), output=320480, compensated lookahead=60.0ms
1.25x: processing=12.29ms for 10s (0.123% realtime), output=192480, compensated lookahead=60.0ms
1.50x: processing=11.56ms for 10s (0.116% realtime), output=160480, compensated lookahead=60.0ms

The 60 ms value is Signalsmith's compensated internal lookahead, not added audible delay; the measured end-to-end processing wall time is shown separately.

Run focused behavior checks:

cargo test --manifest-path desktop/src-tauri/Cargo.toml playback_speed
cargo test --manifest-path desktop/src-tauri/Cargo.toml voice_selection_tests

[johnmatthewtennant]

🤖 ### Speech playback speed

The global generated-speech control appears in Appearance, defaults to 1.00x, and exposes a reset action independently of Pocket voice settings.

[johnmatthewtennant]

🤖 ### Voice settings

Global pitch-preserving playback speed is layered beside the Pocket voice controls.

[johnmatthewtennant]

🤖 New review-intent sweep completed on the frozen stacked candidate.

• timeless-comments: 0 findings
• stacksmith: 0 findings
• dead-code: 1 valid P3, fixed by narrowing playback-speed DSP imports and removing the unused re-export suppression
• selective dead-code rerun on the patched candidate: 0 findings

Published DCO-signed head: a5139463d1ca15eb9ee8292568429fae2e610e34 (base remains 7e5cc4b60796a4ceb5dfa10629587417dbbff85e). Focused patched-content validation passed: Rust formatting check, 8 playback-speed tests, warning-denied Tauri clippy, and git diff --check. GitHub CI finished green with 18 passing and 7 skipped checks. PR remains open and draft; no readiness or merge action was taken.

Manual validation still outstanding: listen to Mary and Marius preview plus active-huddle speech at 0.75x, 1.25x, and 1.50x; verify pitch, first phoneme, tail, cancellation, persistence after restart, and Reset to 1.00x.