diff --git a/10-systems/hosts/gprint.md b/10-systems/hosts/gprint.md index fd258df..f02a489 100644 --- a/10-systems/hosts/gprint.md +++ b/10-systems/hosts/gprint.md @@ -6,7 +6,7 @@ owner: gmikcon status: active confidence: 100 -last_reviewed: 2026-09-29 +last_reviewed: 2026-10-05 review_interval: 90d source_of_truth: ssh-host --- @@ -21,9 +21,9 @@ ## Technical details - **OS:** Ubuntu 24.04.1 LTS, x86_64, 4 cores, 1.9 GiB RAM -- **Disk:** 13G total, ~6.8G used (57%) after cleanup (was 90%) +- **Disk:** 13G total, ~7.0G used (58%) as of 2026-10-05 - **User:** gprint; SSH and sudo credentials — see gnexus-creds -- **Uptime:** long-lived (46d as of 2026-09-29); kernel update pending reboot — safe to reboot, all services in autostart +- **Uptime:** long-lived; kernel update pending reboot — safe to reboot, all services in autostart ## Services - **NFS share** (ports 111, 2049) — primary role, untouched by cleanup @@ -43,6 +43,15 @@ - `README.md` — full node documentation (80 lines, includes Termux + notifications sections) - TWRP beyond0lte 3.7.0_9-2 at `/home/gprint/twrp-beyond0lte.img` (sha256 verified) +## Piper TTS (~/pixel3a-work/tts/) +Installed 2026-10-05 for Navi voice replies (Russian female voice, Telegram voice messages). +- **Binary:** `~/pixel3a-work/tts/piper` (release 2023.11.14-2, linux x86_64 glibc; bundled espeak-ng + libonnxruntime in same dir) +- **Voice:** `~/pixel3a-work/tts/irina.onnx` + `irina.onnx.json` — ru_RU-irina-medium (female, 22050 Hz, ~63 MB, from rhasspy/piper-voices v1.0.0) +- **Test files:** `test.wav` (327 KB, 7.4 s) and `test.ogg` (29 KB, 7.4 s, Opus 32k mono 48 kHz — Telegram-compatible) +- **Usage:** `echo 'текст' | ~/pixel3a-work/tts/piper --model ~/pixel3a-work/tts/irina.onnx --output_file out.wav` +- **Convert for Telegram:** `ffmpeg -i out.wav -c:a libopus -b:a 32k -ar 48000 -ac 1 out.ogg` +- **Performance:** voice loads in ~1.8 s; RTF ~1.55 on this host (7 s audio ≈ 11 s synthesis) — fine for short phrases, RAM ~1.5 GB peak, fits in 1.9 GiB + ## Termux access (on the S10e) - SSH from gprint to the phone: `ssh -i ~/.ssh/s10e_key -p 8022 navi@192.168.1.22` (key `~/.ssh/s10e_key`, ed25519 gprint-navi) - Details and gotchas — see `10-systems/hardware/s10e-navi-node.md` @@ -57,3 +66,4 @@ - scrcpy 1.25: no `--no-audio`/`--time-limit` flags (those are 2.x); headless via `xvfb-run -a scrcpy -s 192.168.1.22:5555 --max-size 800` - With both USB and Wi-Fi adb connected, always use `adb -s ` - sudo over non-interactive ssh: `echo | sudo -S ` (no TTY) +- Piper synthesis on this host is slower than real-time (RTF ~1.55) — use timeouts ≥120 s for long texts diff --git a/90-maintenance/pending-changes/20261005170843-a2ab81dd.json b/90-maintenance/pending-changes/20261005170843-a2ab81dd.json new file mode 100644 index 0000000..b04ebf3 --- /dev/null +++ b/90-maintenance/pending-changes/20261005170843-a2ab81dd.json @@ -0,0 +1,19 @@ +{ + "id": "20261005170843-a2ab81dd", + "kind": "doc", + "target": "10-systems/hosts/gprint.md", + "summary": "Add Piper TTS section to gprint host doc (binary, irina voice, usage, performance)", + "reason": "Piper TTS installed on gprint host 2026-10-05 (verified via ssh-host): binary, ru_RU-irina-medium voice, test wav/ogg files in ~/pixel3a-work/tts/. Documented usage commands and performance characteristics for future voice-reply work.", + "payload": { + "content": "---\ntitle: gprint\ntype: host\nip: 192.168.1.131\npurpose: NFS network share + S10e Android node host (former 3D-printer brain)\nowner: gmikcon\nstatus: active\nconfidence: 100\nlast_reviewed: 2026-10-05\nreview_interval: 90d\nsource_of_truth: ssh-host\n---\n\n# gprint (192.168.1.131)\n\n## Overview\nSmall physically separate x86_64 server in the home LAN. Formerly the brain\nof a 3D printer (Klipper/Moonraker stack) — cleaned up 2026-09-29. Now serves\nan NFS network share and hosts the S10e Android Navi node\n(see `10-systems/hardware/s10e-navi-node.md`).\n\n## Technical details\n- **OS:** Ubuntu 24.04.1 LTS, x86_64, 4 cores, 1.9 GiB RAM\n- **Disk:** 13G total, ~7.0G used (58%) as of 2026-10-05\n- **User:** gprint; SSH and sudo credentials — see gnexus-creds\n- **Uptime:** long-lived; kernel update pending reboot — safe to reboot, all services in autostart\n\n## Services\n- **NFS share** (ports 111, 2049) — primary role, untouched by cleanup\n- **nginx** (port 80) — fluidd site removed; default config\n- **iperf3** (port 5201)\n- **ssh** (port 22)\n- **adb/fastboot/scrcpy** + udev rules (51-android-nexus.rules, vendor 18d1/04e8) — for the S10e node\n\n## S10e node stack (~/pixel3a-work/)\n- `s10e.sh` — control script: screen/tap/swipe/text/key/shell/launch/pkgs/info/lock/unlock/reboot/status/scrcpy\n- `check.sh` — adb connection status\n- `keepalive.sh` + systemd `pixel3a-keepalive.service` (active) — reconnects adb-over-WiFi, re-enables tcpip via USB after phone reboot\n- `pixel3a-watch.service` (active) — USB presence watcher (watch.log)\n- `pixel3a-setup.service` — disabled, one-shot setup completed 8/8\n- `notif-watcher.sh` + systemd `notif-watcher.service` (active, enabled) — polls phone notifications every 10s, new ones appended to `notif.log` (ts|pkg|title|text), dedup in `notif.state`\n- `.pin` — phone PIN (chmod 600)\n- `README.md` — full node documentation (80 lines, includes Termux + notifications sections)\n- TWRP beyond0lte 3.7.0_9-2 at `/home/gprint/twrp-beyond0lte.img` (sha256 verified)\n\n## Piper TTS (~/pixel3a-work/tts/)\nInstalled 2026-10-05 for Navi voice replies (Russian female voice, Telegram voice messages).\n- **Binary:** `~/pixel3a-work/tts/piper` (release 2023.11.14-2, linux x86_64 glibc; bundled espeak-ng + libonnxruntime in same dir)\n- **Voice:** `~/pixel3a-work/tts/irina.onnx` + `irina.onnx.json` — ru_RU-irina-medium (female, 22050 Hz, ~63 MB, from rhasspy/piper-voices v1.0.0)\n- **Test files:** `test.wav` (327 KB, 7.4 s) and `test.ogg` (29 KB, 7.4 s, Opus 32k mono 48 kHz — Telegram-compatible)\n- **Usage:** `echo 'текст' | ~/pixel3a-work/tts/piper --model ~/pixel3a-work/tts/irina.onnx --output_file out.wav`\n- **Convert for Telegram:** `ffmpeg -i out.wav -c:a libopus -b:a 32k -ar 48000 -ac 1 out.ogg`\n- **Performance:** voice loads in ~1.8 s; RTF ~1.55 on this host (7 s audio ≈ 11 s synthesis) — fine for short phrases, RAM ~1.5 GB peak, fits in 1.9 GiB\n\n## Termux access (on the S10e)\n- SSH from gprint to the phone: `ssh -i ~/.ssh/s10e_key -p 8022 navi@192.168.1.22` (key `~/.ssh/s10e_key`, ed25519 gprint-navi)\n- Details and gotchas — see `10-systems/hardware/s10e-navi-node.md`\n\n## Cleanup history (2026-09-29)\n- Removed arm-none-eabi cross-compiler (2.4G), trimmed journald to 100M limit\n- Klipper/Moonraker remains backed up to `/root/klipper-backup-*.tar.gz`, then removed\n- Dead nginx fluidd site removed (nginx -t OK, reloaded); NFS share preserved\n\n## Gotchas\n- USB port 1-2 is glitchy (EMI-disable, connect/disconnect cycle) — keep the phone in port 1-3\n- scrcpy 1.25: no `--no-audio`/`--time-limit` flags (those are 2.x); headless via `xvfb-run -a scrcpy -s 192.168.1.22:5555 --max-size 800`\n- With both USB and Wi-Fi adb connected, always use `adb -s `\n- sudo over non-interactive ssh: `echo | sudo -S ` (no TTY)\n- Piper synthesis on this host is slower than real-time (RTF ~1.55) — use timeouts ≥120 s for long texts" + }, + "status": "applied", + "created_at": "2026-10-05T17:08:43Z", + "updated_at": "2026-10-05T17:08:45Z", + "applied_at": "2026-10-05T17:08:45Z", + "validation": { + "status": "ok", + "issue_count": 0, + "issues": [] + } +}