diff options
| author | salmonik <salmonik@tuta.io> | 2026-11-04 01:33:33 +0100 |
|---|---|---|
| committer | salmonik <salmonik@tuta.io> | 2026-11-04 01:33:33 +0100 |
| commit | 97a26ec4e64a98a24dad7969337d2569339db9a7 (patch) | |
| tree | ada17e2b1f50d053a6bfe4067296e06de848ac87 /CONTEXT.md | |
| parent | config: harden the shared temp directory (diff) | |
| download | senator-sperma-97a26ec4e64a98a24dad7969337d2569339db9a7.tar.gz senator-sperma-97a26ec4e64a98a24dad7969337d2569339db9a7.zip | |
docs: sync with current behavior
Diffstat (limited to 'CONTEXT.md')
| -rw-r--r-- | CONTEXT.md | 13 |
1 files changed, 7 insertions, 6 deletions
| @@ -10,8 +10,8 @@ holds no implementation decisions or specs (those live in `docs/adr/`). | |||
| 10 | Despite the name, it contains **no audio** — only timestamps and durations. | 10 | Despite the name, it contains **no audio** — only timestamps and durations. |
| 11 | - **Recording** — capturing the *audio* of a voice channel into a rolling | 11 | - **Recording** — capturing the *audio* of a voice channel into a rolling |
| 12 | in-memory buffer while the bot is connected. Distinct from *voice activity* | 12 | in-memory buffer while the bot is connected. Distinct from *voice activity* |
| 13 | (metadata). A bot "records" continuously while in a VC (unless opted out) and | 13 | (metadata). Recording is opt-in via `RECORDING_ENABLED` (default false); when |
| 14 | "flushes" on `/record`. | 14 | enabled a bot records continuously while in a VC and "flushes" on `/record`. |
| 15 | - **Window** — the retention horizon of the recording buffer: how many seconds | 15 | - **Window** — the retention horizon of the recording buffer: how many seconds |
| 16 | of the recent past are kept. Older audio is discarded as new audio arrives. | 16 | of the recent past are kept. Older audio is discarded as new audio arrives. |
| 17 | Configured via `RECORDING_WINDOW_SECONDS` (default 900 = 15 minutes). | 17 | Configured via `RECORDING_WINDOW_SECONDS` (default 900 = 15 minutes). |
| @@ -40,10 +40,11 @@ holds no implementation decisions or specs (those live in `docs/adr/`). | |||
| 40 | at the model's native sample rate (typically 22050 Hz mono), resampled to | 40 | at the model's native sample rate (typically 22050 Hz mono), resampled to |
| 41 | 48 kHz stereo via inline linear interpolation before playback. Songbird does | 41 | 48 kHz stereo via inline linear interpolation before playback. Songbird does |
| 42 | not resample — playing un-resampled audio results in wrong playback speed. | 42 | not resample — playing un-resampled audio results in wrong playback speed. |
| 43 | - **TtsState** — struct held in poise `Data`: `OnceCell<Mutex<Piper>>` for the | 43 | - **TtsState** — struct held in poise `Data`: `OnceLock<Arc<std::sync::Mutex<Piper>>>` |
| 44 | model instance, `DashMap<String, (Vec<u8>, Instant)>` for the text→audio cache. | 44 | for the model instance, `DashMap<String, (Vec<u8>, Instant)>` for the |
| 45 | Piper loaded lazily on first TTS call. Cache keyed by full phrase string, | 45 | text→audio cache. Piper is loaded lazily on first TTS call, off the async |
| 46 | evicted by TTL. No Google TTS fallback. | 46 | runtime. Cache is keyed by the full phrase, bounded by entries/bytes, and |
| 47 | evicted by TTL. No external TTS fallback. | ||
| 47 | - **Pre-generation** — on first voice event for a user, synthesize all 6 event | 48 | - **Pre-generation** — on first voice event for a user, synthesize all 6 event |
| 48 | phrases (join/leave/deafen/undeafen/mute/unmute) with their display name and | 49 | phrases (join/leave/deafen/undeafen/mute/unmute) with their display name and |
| 49 | cache them. ~120ms first time, instant after. Lazy — no startup cost. | 50 | cache them. ~120ms first time, instant after. Lazy — no startup cost. |