| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Replace PC-98 landing aesthetic with Heisei Twilight theme
Layered new heisei.css (loaded after pc98.css) over the public landing
page so the late-2000s/early-2010s anime cue lands subtly without a
ground-up rewrite. Internal VN/contracts/daemons screens are untouched.
What changed
- Twilight gradient backdrop (deep night → dusk plum → horizon rose +
amber) replaces the chunky animated GIF background.
- Soft bokeh plates and a faint scanline overlay sit behind the content
for atmosphere.
- Floating header is now frosted glass with a 1px hairline; the system
status pills use Inter-tracked uppercase labels with desaturated
cyan/magenta indicators.
- Vertical kanji masthead becomes a quiet Noto Serif JP column rather
than an arcade box; English subtitle picks up a warm amber tone.
- Issue badge is a flat pill; LED heart still pulses but in muted rose.
- Empty hero state shows a faint tactical reticle (corner brackets) and
bilingual JP/EN tagline — Ghost-in-the-Shell SAC HUD vibe.
- Mission and MAKIMA panels render as Heisei magazine columns with a
Noto Serif JP headline, accent rule (amber for Mission, magenta for
MAKIMA), and Inter body copy.
- Buttons are flat with hairline borders and a left-edge amber accent
sweep on hover; the Login CTA keeps a warm gradient as the primary
action.
- Bottom Velocity/Energy strip becomes a small HUD pill with corner
bracket detailing.
- Mobile breakpoint tightens header, padding, headline size, and stacks
the secondary CTAs back to a column.
- prefers-reduced-motion respected for the heart pulse and image pan.
How to verify
- cd frontend && npm install && npm run dev
- Compare http://localhost:5173 with the live soryu.co (PC-98 build)
- Toggle the Mission and MAKIMA panels; resize to 390x844 for mobile.
Implementation notes
- Single new file: frontend/src/styles/heisei.css (~16 KB).
- main.tsx imports heisei.css after pc98.css so the override wins.
- No JSX/component changes — design lives entirely in CSS, easy to
revert by removing the import line.
- npm run build passes (vite v5.4, 91 kB CSS gzipped to 17.8 kB).
* Attach before/after screenshots for review
Six viewport/state combinations captured at 2x DPR with Playwright
(desktop 1440x900, mobile 390x844). before-* show the live
soryu.co PC-98 build; after-* show the heisei-twilight branch
served by vite dev. Drop this commit before merge if you'd
rather not carry the assets in tree.
* Move CTA row into a top navbar below the floating header
Per review feedback, lift Mission / MAKIMA / Login from the bottom of
the cover into a fixed navbar that sits flush beneath the floating
header — standard site pattern.
- .cta-area is now position: fixed, top: 72px, height 52px (top: 56px,
height 48px on mobile), with the same frosted-glass + hairline
treatment as the header so they read as one stacked nav block.
- Buttons become ghost-link items inside the bar with an animated
underline-from-center on hover; Login keeps its warm amber→rose
primary treatment but compact.
- cover-content top padding bumped to 156px (desktop) / 124px (mobile)
so the masthead and hero clear both stripes.
- bottom-stats no longer needs to dodge a fixed-bottom CTA — pulled
back to bottom: 24px (desktop) and ~20px + safe-area (mobile).
- npm run build passes (CSS 91.6 kB / 17.9 kB gzip).
Screenshots in docs/heisei-screenshots/after-*.jpg refreshed against
the new navbar layout (1440x900 desktop, 390x844 mobile, 2x DPR).
* Center mobile navbar buttons (kill leftover pc98 margin-top)
The bottom-CTA layout in pc98.css had a mobile-specific
margin-top: 20px on .taisho-cta, which carried over into the new
top navbar and pushed every button 20px below the navbar's
vertical center on mobile (visible misalignment at 390x844 — the
Mission/MAKIMA/Login pills were sitting near the bottom edge of
the bar).
- Add margin: 0 !important to .taisho-cta in heisei.css (and to
the mobile-block override) so the legacy column-layout margin
is wiped on both breakpoints.
- Switch buttons to height: 32px + line-height: 32px for stable
vertical centering inside the 48px (mobile) / 52px (desktop)
navbar.
- Force align-self: center on .cta-left/.cta-right so they don't
inherit any stretched height from previously-applied rules.
Verified via Playwright DOM probe at 390x844:
navbar y=56–104 (centerY 80) · buttons y=64–96 (centerY 80) ✓
And at 1280x720:
navbar y=72–124 (centerY 98) · buttons y=83–113 (centerY 98) ✓
Screenshots in docs/heisei-screenshots/after-*.jpg refreshed.
* Tighten mission/makima top space and use highlighted nav state
The CTA row is now a top navbar, so the old 'Close' label on the
active panel button is redundant — the button itself should look
like the current page. Two fixes here:
1. Replace 'Close' with a persistent active highlight.
- LandingPage.tsx: button keeps its label ('Mission' / 'MAKIMA')
and gains an is-active class + aria-current="page" when its
panel is open. heisei.css renders the active state as a solid
amber underline + amber text — same visual language as the
hover indicator, just persistent.
- The button still toggles the panel closed when clicked again,
so the interaction is unchanged; only the labelling moved.
2. Remove the white-space band above mission content.
- The legacy .modern-landing-page wrapper from pc98.css pinned
itself at top: 120px and height: calc(100vh - 120px), matching
the original 120px-tall floating header. The new header is
72px (56px on mobile), which left ~50px of dead space below
the navbar before the mission/makima content started.
- Override the wrapper to top: 72px / height: calc(100vh - 72px)
on desktop and 56px on mobile so it sits flush under the
header.
- Cover-content padding-top retuned for the new wrapper origin:
desktop default 84px, mission-mode 72px (navbar 52 + 20 gap).
Mobile default 76px, mission-mode 64px (navbar 48 + 16 gap).
Verified at 1280x720:
navbar y=72-124 · mission headline starts y=144 · 20px gap ✓
active Mission button: 'taisho-cta is-active', aria-current=page ✓
npm run build passes (CSS 92.6 kB / 18.1 kB gzip). Screenshots in
docs/heisei-screenshots/after-*.jpg refreshed against the tightened
layout.
* Align mission/makima body, fix mobile bg, add typewriter hero rotator
Three fixes coming out of QA on staging (soryu.eirin.xyz):
1. Mission / MAKIMA body alignment
- The amber/magenta accent rule + 18px indent used to live on
.mission-headline only, so paragraphs (and the mission image)
started 18px to the LEFT of the headline.
- Move the rule + padding-left onto the .mission-screen wrapper
itself. Headline, image, and all paragraph(s) now share one
consistent column edge.
- Tighten the makima-badge so it doesn't stretch the grid column
(justify-self: start + width: max-content). Image gets explicit
width: 100%; display: block to behave inside the column.
2. Mobile background coverage
- pc98.css mobile-block override ('@media max-width:768px') reset
.modern-landing-page to position: relative + margin-top: 100px,
leaving a 100-156px band at the top where only the body color
showed and pushing the wrapper bottom past the viewport.
- Force position: fixed on the wrapper at every breakpoint with
!important. Use 100dvh (dynamic viewport height) so the gradient
follows iOS Safari's collapsing URL bar rather than getting
clipped.
- Mirror the dusk gradient onto <body> as a fallback so any
sub-pixel rounding gap shows the same colour, not white.
3. Typewriter hero rotator
- New <TypewriterRotator/> component (frontend/src/components/
TypewriterRotator.tsx) — types, holds, deletes, gaps, advances.
Honours prefers-reduced-motion (renders the first phrase static).
- LandingPage.tsx renders it inside a new .hero-tagline wrapper
where the old static CSS ::after lived. Five phrases, all
bilingual JP/EN in the same Heisei tactical voice:
* 低遅延ストリーミング · LOW-LATENCY OBSERVABILITY
* リアルタイム監視 · REAL-TIME SURVEILLANCE
* ミッションクリティカル · MISSION-CRITICAL INFRASTRUCTURE
* エンドツーエンド可視化 · END-TO-END VISIBILITY
* 安全な意思決定 · SECURE DECISIONS AT THE EDGE
- Hide the old .hero::after pseudo-tagline; .hero-tagline keeps
the same centred position, amber colour, and tracking.
- Caret blink animation (steps(1), 1s) — separate from the typing
timing so it always blinks at human speed.
npm run build passes (CSS 94 kB / 18.2 kB gzip, JS 238 kB / 75 kB
gzip — +1.5 kB net for the rotator + new phrases).
Verified at 1280x720 / 390x844 via Playwright DOM probe:
- mission/makima: headline.x === paragraph.x === image.x ✓
- mobile: wrapper y=56, h=788, bottom=844 (full coverage) ✓
- typewriter: types, holds, deletes, advances to phrase 2 ✓
Screenshots in docs/heisei-screenshots/after-*.jpg refreshed.
Staging at soryu.eirin.xyz updated via rsync.
* Center hero on homepage, image-left text-right on mission/makima
- Default homepage now uses single-column grid; the SORYU masthead is
absolutely positioned in the upper-left of the cover frame, so the
hero typewriter is genuinely centered on the page (verified
taglineCenterX === pageCenter at 1280px).
- Mission/MAKIMA on desktop (>=1025px) switch to a flex layout with
the image absolutely positioned in a 360px left gutter and all body
text (badge -> headline -> paragraphs) flowing on the right.
- Mobile and tablet (<=1024px) keep the stacked layout; masthead
returns to inline flow under the navbar.
- MAKIMA badge uses flex `order: -1` so it appears above the headline
visually, even though the JSX renders headline first.
* Replace PC-98 character image with tactical observability mesh SVG
- New /public/mission-tactical.svg (~9KB, license-clean, hand-crafted):
amber wireframe globe + magenta mesh links, GitS:SAC-style HUD chrome
(corner brackets, scanlines, telemetry strip, NODE-001 TYO label,
pulsing origin node), kanji watermark.
- Uses CSS palette vars (amber #e8b87a / magenta #d96a8a / cyan #9ad7e0).
- .mission-image switched to object-fit:contain with a dark fill so the
vector renders fully and never crops; pc98 mission-pan animation is
killed for the SVG path.
* Animate mission tactical SVG: flowing data along mesh links
- Each of 6 mesh arcs now has an id (#arc1..#arc6) with two layered
animations: a flowing stroke-dashoffset (so the lines themselves
appear to ripple), and a separate breathing opacity cycle.
- 8 packet circles ride the arcs via <animateMotion> + <mpath>, with
staggered begin times (0s..2.0s) and varied durations (2.6s..4.0s)
so traffic feels asynchronous, not metronomic.
- Amber arcs carry amber packets, magenta arcs carry magenta packets.
- Hub nodes now have layered radial pulses (filled core + expanding
ring) on independent timings; cyan relay dots blink gently.
- REC indicator pulses 1.4s.
All animations are SVG-native (no JS, no external library); browsers
respect prefers-reduced-motion via UA defaults on <animate>.
* chore: gitignore tsconfig.tsbuildinfo (auto-generated incremental build cache)
* Drop mission-image ring border; reskin Login hover (no glow)
- The animated SVG already draws its own hairline frame + corner
brackets at the viewBox edge. Remove the `box-shadow: 0 0 0 1px`
ring (in both the base `.mission-image` rule and the
desktop-min-1025px `.mission-screen .mission-image` override) so
the image isn't double-framed. Drop-shadow for elevation kept.
- Replace the Login button hover effect. The amber drop-shadow halo
used the same visual language as the active Mission/MAKIMA nav
buttons (which also glow amber underneath); on hover Login now
inverts to a deep-night background with amber ink and a hairline
::after rule that slides in from the left along the top edge —
reads as 'armed' rather than 'glowing'. Letter-spacing widens
slightly on hover (0.18em → 0.22em) for a subtle ink-spread.
Added :active and :focus-visible states for keyboard a11y.
* Remove vertical accent rule from Mission/MAKIMA text columns
The 2px amber/magenta border-left + 18px indent on the right-column
text was reading as another border running parallel to the SVG image
on the left. The animated SVG already supplies all the framing the
panels need, so the rule is now redundant noise.
Removed in three places:
- .mission-screen wrapper (base + .makima-screen variant)
- desktop (>=1025px) per-block rule on headline + paragraph
- mobile (<=768px) leftover padding-left: 14px on headline
Verified: borderLeft = 0 on wrapper, headline, and paragraph(s) on
both Mission and MAKIMA. Headline + body now sit flush with the badge
at the same column edge (x=680 at 1280px).
* Mission/MAKIMA backdrop: solid dark plate (no horizon bleed)
The panel backdrop was rgba(11,17,36,0.62→0.85) — semi-transparent —
so the page's dusk-pink horizon at the bottom of the gradient bled
through behind the panel content, drawing the eye to the page edge
and making the background gradient stand out instead of disappearing.
Switched to a near-opaque #0a1024 → #0b1530 vertical gradient (slight
slope kept for depth) and bumped backdrop-filter to blur(20px)
saturate(80%) so any residual ambient is fully diffused. The panel
now reads as its own surface rather than a tinted overlay.
Verified: bottom-center pixel is rgb(10,20,47) — uniform deep night
across the whole panel, no pink bleed at the page bottom.
* Drop dark rounded plate behind Mission/MAKIMA images
The .mission-image and .makima-logo had a semi-opaque dark fill
(rgba(11,17,36,0.55-0.6)) plus border-radius: 2px. Combined with
object-fit: contain (which letterboxes the SVG inside the larger
360px box), the dark fill was visible as a rounded rectangle
'frame' around the artwork — most noticeable on MAKIMA where the
small circular logo sits in a much larger box.
Switched both to transparent background + border-radius: 0 so the
artwork floats directly on the panel backdrop. Drop-shadow kept on
both for elevation; .makima-logo's amber halo + hairline ring
removed too (it was painting the same noise).
Verified by sampling pixels at the four image corners + inner mid:
all read as the same uniform deep-night as the panel backdrop
(rgb(10, 14-18, 30-42)). No plate, no rounded edge.
* Remove box-shadow halo around Mission/MAKIMA image bounding boxes
* Mission/MAKIMA backdrop: slightly translucent (alpha 0.72→0.84, blur 28px)
---------
Co-authored-by: soryu-co <bot@soryu.co>
|
| |
|
|
|
|
|
|
|
|
|
| |
* feat: soryu-co/soryu - makima: Remove aarch64-unknown-linux-gnu from release workflow
* WIP: heartbeat checkpoint
* WIP: heartbeat checkpoint
* WIP: heartbeat checkpoint
* feat: soryu-co/soryu - makima: Implement non-blocking ask with client-side polling for reconcile mode
|
| |
|
|
|
|
|
|
|
|
|
| |
* feat: soryu-co/soryu - makima: Update install.sh to point at public makima repo
* feat: soryu-co/soryu - makima: Update daemons page download links to use public makima releases
* feat: soryu-co/soryu - makima: Research Cloudflare Containers for running full makima daemon
* feat: soryu-co/soryu - makima: Push container image to ghcr.io/soryu-co/makima namespace
* feat: soryu-co/soryu - makima: Add workflow to sync public files to soryu-co/makima repo
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Analyze the compound engineering plugin (https://github.com/EveryInc/compound-engineering-plugin)
and propose 6 features inspired by its patterns for adoption into makima:
- Multi-agent parallel review system (spawn-group/wait-group)
- Knowledge accumulation / compound learning phase
- Parallel plan deepening with research agents
- Workflow presets / pipeline templates (LFG-style one-command pipelines)
- Structured findings tracking with severity and lifecycle
- Reusable task templates with meta-commands
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Add docs/contract-management-spec.md with full system design
- Add docs/plans/implementation-plan.md with 5-phase rollout plan
- Add validate_deliverable() function and use in mark_deliverable_complete
- Add PhaseChangeResult enum and change_contract_phase_with_version() with FOR UPDATE locking
- Enforce phase_guard at API level for all callers
This addresses critical issues in contract management:
- Deliverable validation to prevent marking non-existent deliverables complete
- Version conflict detection for phase changes with row locking
- Phase guard enforcement at API level (applies to all callers including supervisors)
- Comprehensive specification and implementation plan for future phases
|
|
|
* Task completion checkpoint
* Task completion checkpoint
* Task completion checkpoint
* Add Qwen3-TTS research document for live TTS replacement
Research findings for replacing Chatterbox TTS with Qwen3-TTS-12Hz-0.6B-Base:
- Current TTS: Chatterbox-Turbo-ONNX with batch-only generation, no streaming
- Qwen3-TTS: 97ms end-to-end latency, streaming support, 3-second voice cloning
- Voice cloning: Requires 3s reference audio + transcript (Makima voice planned)
- Integration: Python service with WebSocket bridge (no ONNX export available)
- Languages: 10 supported including English and Japanese
Document includes:
- Current architecture analysis (makima/src/tts.rs)
- Qwen3-TTS capabilities and requirements
- Feasibility assessment for live/streaming TTS
- Audio clip requirements for voice cloning
- Preliminary technical approach with architecture diagrams
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* [WIP] Heartbeat checkpoint - 2026-01-27 03:11:15 UTC
* Add Qwen3-TTS research documentation
Comprehensive research on replacing Chatterbox TTS with Qwen3-TTS-12Hz-0.6B-Base:
- Current TTS implementation analysis (Chatterbox-Turbo-ONNX in makima/src/tts.rs)
- Qwen3-TTS capabilities: 97ms streaming latency, voice cloning with 3s reference
- Cross-lingual support: Japanese voice (Makima/Tomori Kusunoki) speaking English
- Python microservice architecture recommendation (FastAPI + WebSocket)
- Implementation phases and technical approach
- Hardware requirements and dependencies
Key findings:
- Live/streaming TTS is highly feasible with 97ms latency
- Voice cloning fully supported with 0.95 speaker similarity
- Recommended: Python microservice with WebSocket streaming
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add comprehensive Qwen3-TTS integration specification
This specification document defines the complete integration of
Qwen3-TTS-12Hz-0.6B-Base as a replacement for the existing Chatterbox-Turbo
TTS implementation. The document covers:
## Functional Requirements
- WebSocket endpoint /api/v1/speak for streaming TTS
- Voice cloning with default Makima voice (Japanese VA speaking English)
- Support for custom voice references
- Detailed client-to-server and server-to-client message protocols
- Integration with Listen page for bidirectional speech
## Non-Functional Requirements
- Latency targets: < 200ms first audio byte
- Audio quality: 24kHz, mono, PCM16/PCM32f
- Hardware requirements: CUDA GPU with 4-8GB VRAM
- Scalability: 10 concurrent sessions per GPU
## Architecture Specification
- Python TTS microservice with FastAPI/WebSocket
- Rust proxy endpoint in makima server
- Voice prompt caching mechanism (LRU cache)
- Error handling and recovery strategies
## API Contract
- Complete WebSocket message format definitions (TypeScript)
- Error codes and responses (TTS_UNAVAILABLE, SYNTHESIS_ERROR, etc.)
- Session state machine and lifecycle management
## Voice Asset Requirements
- Makima voice clip specifications (5-10s WAV, transcript required)
- Storage location: models/voices/makima/
- Metadata format for voice management
## Testing Strategy
- Unit tests for Python TTS service and Rust proxy
- Integration tests for WebSocket flow
- Latency benchmarks with performance targets
- Test data fixtures for various text lengths
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add Qwen3-TTS implementation plan
Comprehensive implementation plan for replacing Chatterbox-TTS with
Qwen3-TTS streaming TTS service, including:
- Task breakdown with estimated hours for each phase
- Phase 1: Python TTS microservice (FastAPI, WebSocket)
- Phase 2: Rust proxy integration (speak.rs, tts_client.rs)
- Detailed file changes and new module structure
- Testing plan with unit, integration, and latency benchmarks
- Risk assessment with mitigation strategies
- Success criteria for each phase
Based on specification in docs/specs/qwen3-tts-spec.md
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add author and research references to TTS implementation plan
Add links to research documentation and author attribution.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* [WIP] Heartbeat checkpoint - 2026-01-27 03:25:06 UTC
* Add Python TTS service project structure (Phase 1.1-1.3)
Create the initial makima-tts Python service directory structure with:
- pyproject.toml with FastAPI, Qwen-TTS, and torch dependencies
- config.py with pydantic-settings TTSConfig class
- models.py with Pydantic message models (Start, Speak, Stop, Ready, etc.)
This implements tasks P1.1, P1.2, and P1.3 from the Qwen3-TTS implementation plan.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add TTS engine and voice manager for Qwen3-TTS (Phase 1.4-1.5)
Implement core TTS functionality:
- tts_engine.py: Qwen3-TTS wrapper with streaming audio chunk generation
- voice_manager.py: Voice prompt caching with LRU eviction and TTL support
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* [WIP] Heartbeat checkpoint - 2026-01-27 03:30:06 UTC
* Add TTS proxy client and message types (Phase 2.1, 2.2, 2.4)
- Add tts_client.rs with TtsConfig, TtsCircuitBreaker, TtsError,
TtsProxyClient, and TtsConnection structs for WebSocket proxying
- Add TTS message types to messages.rs (TtsAudioEncoding, TtsPriority,
TtsStartMessage, TtsSpeakMessage, TtsStopMessage, TtsClientMessage,
TtsReadyMessage, TtsAudioChunkMessage, TtsCompleteMessage,
TtsErrorMessage, TtsStoppedMessage, TtsServerMessage)
- Export tts_client module from server mod.rs
- tokio-tungstenite already present in Cargo.toml
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add TTS WebSocket handler and route (Phase 2.3, 2.5, 2.6)
- Create speak.rs WebSocket handler that proxies to Python TTS service
- Add TtsState fields (tts_client, tts_config) to AppState
- Add with_tts() builder and is_tts_healthy() methods to AppState
- Register /api/v1/speak route in the router
- Add speak module export in handlers/mod.rs
The handler forwards WebSocket messages bidirectionally between
the client and the Python TTS microservice with proper error handling.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add Makima voice profile assets for TTS voice cloning
Creates the voice assets directory structure with:
- manifest.json containing voice configuration (voice_id, speaker,
language, reference audio path, and Japanese transcript placeholder)
- README.md with instructions for obtaining voice reference audio
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Add Rust-native Qwen3-TTS integration research document
Research findings for integrating Qwen3-TTS-12Hz-0.6B-Base directly into
the makima Rust codebase without Python. Key conclusions:
- ONNX export is not viable (unsupported architecture)
- Candle (HF Rust ML framework) is the recommended approach
- Model weights available in safetensors format (2.52GB total)
- Three components needed: LM backbone, code predictor, speech tokenizer
- Crane project has Qwen3-TTS as highest priority (potential upstream)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* [WIP] Heartbeat checkpoint - 2026-01-27 11:21:43 UTC
* [WIP] Heartbeat checkpoint - 2026-01-27 11:24:19 UTC
* [WIP] Heartbeat checkpoint - 2026-01-27 11:26:43 UTC
* feat: implement Rust-native Qwen3-TTS using candle framework
Replace monolithic tts.rs with modular tts/ directory structure:
- tts/mod.rs: TtsEngine trait, TtsEngineFactory, shared types (AudioChunk,
TtsError), and utility functions (save_wav, resample, argmax)
- tts/chatterbox.rs: existing ONNX-based ChatterboxTTS adapted to implement
TtsEngine trait with Mutex-wrapped sessions for Send+Sync
- tts/qwen3/mod.rs: Qwen3Tts entry point with HuggingFace model loading
- tts/qwen3/config.rs: Qwen3TtsConfig parsing from HF config.json
- tts/qwen3/model.rs: 28-layer Qwen3 transformer with RoPE, GQA (16 heads,
8 KV heads), SiLU MLP, RMS norm, and KV cache
- tts/qwen3/code_predictor.rs: 5-layer MTP module predicting 16 codebooks
- tts/qwen3/speech_tokenizer.rs: ConvNet encoder/decoder with 16-layer RVQ
- tts/qwen3/generate.rs: autoregressive generation loop with streaming support
Add candle-core, candle-nn, candle-transformers, safetensors to Cargo.toml.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: integrate TTS engine into speak WebSocket handler
- Update speak.rs handler to use TTS engine directly from SharedState
instead of returning a stub "not implemented" error
- Add TtsEngine (OnceCell lazy-loaded) to AppState in state.rs with
get_tts_engine() method for lazy initialization on first connection
- Implement full WebSocket protocol: client sends JSON speak/cancel/stop
messages, server streams binary PCM audio chunks and audio_end signals
- Create voices/makima/manifest.json for Makima voice profile configuration
- All files compile successfully with zero errors
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* feat: add /speak TTS page with WebSocket audio playback
Add a new /speak frontend page for text-to-speech via WebSocket.
The page accepts text input and streams synthesized PCM audio through
the Web Audio API. Includes model loading indicator, cancel support,
and connection status. Also adds a loading bar to the listen page
ControlPanel during WebSocket connection.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
|
