feat(transcription): move transcription onto one engine and current models

.claude/skills/verify/SKILL.md

@@ -266,7 +266,9 @@ its profile. It keeps the vault and `.obsidian-e2e-artifacts/`. Delete
   live in `screens/scenes.mjs`, the instance calls in `screens/instance.mjs`,
   and the contact sheet in `screens/sheet.mjs`.
 - `.claude/skills/verify/fixtures.mjs` runs as `feed`, `feed-b`, `transcripts`,
-  `library`, `retitle`, `requests`, `audio [vault path]` or `stop`. `library`
+  `private`, `library`, `retitle`, `requests`, `audio [vault path]` or `stop`.
+  `private` prints a feed at `priv3.supercast.com`, an address PodNotes keeps in
+  SecretStorage as it does a Supercast subscriber's. `library`
   prints the URL of a manifest of eight podcasts, one whose feed answers 404, and
   one more feed to add. Their episodes' media is silence as long as the feed
   says. `#213` has chapters and a WebVTT transcript, and `Signal & Noise` has a 10 s video episode that

.claude/skills/verify/features/transcripts.md

@@ -4,8 +4,9 @@ A user transcribes the current episode. When its feed publishes a transcript
 (`<podcast:transcript>`), PodNotes writes it as timestamped paragraphs with no
 API key and no call to OpenAI or Deepgram. Otherwise, and always through
 `Transcribe current episode from audio`, it sends the audio to the configured
-provider with the user's key. The note goes to `Transcript path`, filled
-from `Transcript template`, and opens.
+provider with the user's key. That is Deepgram when `Speaker diarization` is on
+with Deepgram as the provider, and OpenAI otherwise. The note goes to
+`Transcript path`, filled from `Transcript template`, and opens.
 
 ## Sub-features
 
@@ -14,6 +15,8 @@ from `Transcript template`, and opens.
 - `transcript-paid` keeps the audio path for an episode without a published transcript: the key notice without a key, the provider with one.
 - `transcript-failed` shows the fetch failure when a published transcript cannot be loaded, names the audio command when a key is set, and sends nothing to a provider.
 - `transcript-audio` transcribes the audio with the provider through `Transcribe current episode from audio`, even when a published transcript exists.
+- `transcript-openai` sends OpenAI the audio as `gpt-transcribe` uploads of at most 24 MB and 2,070 s, one at a time, and writes paragraphs without timestamps or speaker labels.
+- `transcript-deepgram` writes timestamped paragraphs with numbered speakers. Deepgram gets the address of an episode a saved public feed lists, and the audio itself for a private feed's episode, a local file, and an address that carries a credential or is not public.
 - `transcript-existing` refuses a second run while the note exists.
 
 ## How to get to it (user POV)
@@ -35,6 +38,12 @@ Preconditions:
 - **Key.** A fake key is a precondition, not a user path: `CLI eval code='app.secretStorage.setSecret("podnotes-openai-api-key-424242", "sk-verify-fixture-not-real"), app.plugins.plugins.podnotes.settings.openAISecretId = "podnotes-openai-api-key-424242", app.plugins.plugins.podnotes.saveSettings(), "key set"'`. With it, `Untranscribed Episode` downloads its WAV and `$NET log` shows a `POST https://api.openai.com/v1/audio/transcriptions` answered by the fixture provider.
 - **Failed.** Play `Broken Transcript Episode` and run the command. The progress notice ends with `The published transcript could not be fetched: Network response returned HTTP 404.`, followed with a key by `Run "Transcribe current episode from audio" to transcribe it with OpenAI instead, which uses your API credit.` `$NET log` has the transcript GET (it records no status) and no provider call or `.wav`.
 - **Audio command.** On the same episode, run `UI key Ctrl+p`, `UI type .prompt-input 'Transcribe current episode from audio'` and `UI click .suggestion-item --text 'from audio'`. With a key, the WAV is fetched and OpenAI (the fixture provider) answers.
+- **OpenAI in parts.** Write an MP3 longer than 2,070 s into the vault, for example 70 min in 4 MB: `ffmpeg -f lavfi -i sine=frequency=440:duration=4200 -ac 1 -ar 8000 -b:a 8k "$VAULT/Fixtures/Long Fixture.mp3"`. Play it ([local-files.md](./local-files.md)) and run the audio command. `$NET log` shows three `POST https://api.openai.com/v1/audio/transcriptions` entries of about 1.4 MB each, one after the other, and the note holds the fixture text three times with no timestamps.
+- **Deepgram.** Set a fake Deepgram key as above, with the id `podnotes-deepgram-api-key-424242` and the setting `deepgramSecretId`. In Settings turn on `Speaker diarization` and pick `Deepgram` under `Diarization provider`. The fixture provider answers with two speakers, the second 12 s in, so the note reads `[00:00:00](...) **1:** This text came from the fixture provider.` and `[00:00:12](...) **2:** A second speaker answers.`
+    - Public. Play `Fixture Episode 1` of `PodNotes Fixture` and run the audio command. `$NET log` shows one `POST https://api.deepgram.com/v1/listen` with `contentType` `application/json` and under 100 `uploadBytes`, and no request for `/episode-1.wav` after the command.
+    - Private. `FEED=$(node .claude/skills/verify/fixtures.mjs private)` prints a feed at `priv3.supercast.com`. Add `PodNotes Private` and check that `data.json` saved it with a `urlSecretId`. Play its `Fixture Episode 1` and run the audio command. `$NET log` shows a `GET` for `/p/episode-1.wav`, then one `POST https://api.deepgram.com/v1/listen` with `contentType` `audio/wav` and the WAV's 960,044 `uploadBytes`.
+    - Local. Play `Fixtures/Local Fixture.wav` and run the audio command. The `POST` carries `audio/wav`, with no `GET` before it.
+- **Speaker labels with OpenAI.** With `Speaker diarization` on and `OpenAI (no speaker labels)` as the provider, the progress notice ends with `Saved without speaker labels, because OpenAI does not label speakers. These providers do: Deepgram.` and stays until clicked.
 - **Proof.** Screenshot the opened note (`npm run capture:e2e-obsidian -- screenshot .obsidian-e2e-artifacts/transcripts/note.png`), keep the note text, `$NET log` and `fixtures.mjs requests` for every run, and read `data.json` to show no new field.
 
 ## Gotchas
@@ -44,6 +53,6 @@ Preconditions:
 
 - `net.mjs` sees `window.fetch` and `requestUrl`. The player streams audio through the media element, which only `fixtures.mjs requests` shows.
 - A secret id must be `podnotes-openai-api-key` or `podnotes-openai-api-key-<n>`. Any other id reads as missing on this device.
-- Notices hide after about 5 s. Read them right after the command. They also cover the PodNotes view's head and the full player's collapse button, and `ui.mjs` refuses a covered click, so dismiss them first.
+- The progress notice hides about 5 s after a transcript is saved. A failure, or a transcript saved with a warning, stays until clicked. Notices also cover the PodNotes view's head and the full player's collapse button, and `ui.mjs` refuses a covered click, so dismiss them first.
 - After editing `fixtures.mjs`, run `fixtures.mjs stop` before `fixtures.mjs transcripts`. A running server keeps serving the old routes.
 - Every fixture transcript cue falls inside the 60 s WAV, so its links seek within the audio.

.claude/skills/verify/fixtures.mjs

@@ -5,6 +5,8 @@
 //   node .claude/skills/verify/fixtures.mjs feed           start or reuse the feed server, print the feed URL
 //   node .claude/skills/verify/fixtures.mjs feed-b         same server, print a second feed whose episode titles repeat the first's
 //   node .claude/skills/verify/fixtures.mjs transcripts    same server, print a feed whose episodes publish transcripts
+//   node .claude/skills/verify/fixtures.mjs private        same server, print a feed at an address PodNotes keeps
+//                                                          private, as a Supercast subscriber's feed is
 //   node .claude/skills/verify/fixtures.mjs library        same server, print the URL of a JSON manifest of eight
 //                                                          podcasts as PodNotes saves them (Night Owl Science's feed
 //                                                          answers 404) and one more feed to add. Each episode's
@@ -26,7 +28,7 @@ import { Readable, pipeline } from "node:stream";
 import { fileURLToPath } from "node:url";
 import { promisify } from "node:util";
 import { loadRunnerConfig, resolveProvisionOptions } from "obsidian-e2e/runner";
-import { FIXTURE_HOST } from "../../../scripts/e2e-instance.mjs";
+import { FIXTURE_HOST, SUPERCAST_FIXTURE_HOST } from "../../../scripts/e2e-instance.mjs";
 
 const root = fileURLToPath(new URL("../../..", import.meta.url));
 // Stable per worktree, so a feed saved in the vault keeps working across restarts.
@@ -36,11 +38,14 @@ const base = `http://${host}:${port}`;
 const feedUrl = `${base}/feed.xml`;
 const feedBUrl = `${base}/feed-b.xml`;
 const transcriptsUrl = `${base}/transcripts.xml`;
+const PRIVATE_ROUTE = "/feeds/verify-subscriber-token/feed.xml";
+const privateUrl = `http://${SUPERCAST_FIXTURE_HOST}:${port}${PRIVATE_ROUTE}`;
 const libraryUrl = `${base}/library.json`;
 const local = `http://127.0.0.1:${port}`;
 const FEED_TITLE = "PodNotes Fixture";
 const FEED_B_TITLE = "PodNotes Fixture B";
 const TRANSCRIPTS_TITLE = "PodNotes Transcripts";
+const PRIVATE_TITLE = "PodNotes Private";
 const transcriptFixtures = path.join(root, "tests/fixtures/transcripts");
 // Media made here outlives the server; node_modules/.cache is where tools keep caches.
 const CACHE = path.join(root, "node_modules/.cache/podnotes-fixtures");
@@ -548,6 +553,13 @@ function serve() {
 		"/b/artwork.svg": ["image/svg+xml", Buffer.from(artwork(FEED_B_TITLE, "#0f766e"))],
 		"/b/episode-1.wav": ["audio/wav", audio],
 		"/b/episode-2.wav": ["audio/wav", audio],
+		[PRIVATE_ROUTE]: [
+			"application/rss+xml",
+			Buffer.from(feedXml(audio.length, PRIVATE_TITLE, "/p")),
+		],
+		"/p/artwork.svg": ["image/svg+xml", Buffer.from(artwork(PRIVATE_TITLE, "#be123c"))],
+		"/p/episode-1.wav": ["audio/wav", audio],
+		"/p/episode-2.wav": ["audio/wav", audio],
 		"/transcripts.xml": ["application/rss+xml", Buffer.from(transcriptsFeedXml(audio.length))],
 		"/t/artwork.svg": ["image/svg+xml", Buffer.from(artwork("Transcripts", "#b45309"))],
 		"/t/buzzcast.html": ["text/html; charset=utf-8", transcript("buzzcast.html")],
@@ -748,14 +760,15 @@ try {
 	else if (verb === "feed") await startFeed(feedUrl);
 	else if (verb === "feed-b") await startFeed(feedBUrl);
 	else if (verb === "transcripts") await startFeed(transcriptsUrl);
+	else if (verb === "private") await startFeed(privateUrl);
 	else if (verb === "library") await startFeed(libraryUrl);
 	else if (verb === "retitle") await retitleFeed();
 	else if (verb === "requests") await printRequests();
 	else if (verb === "audio") await writeAudio(arg);
 	else if (verb === "stop") await stopFeed();
 	else
 		throw new Error(
-			"Usage: fixtures.mjs feed | feed-b | transcripts | library | retitle | requests | audio [vault-relative path] | stop",
+			"Usage: fixtures.mjs feed | feed-b | transcripts | private | library | retitle | requests | audio [vault-relative path] | stop",
 		);
 } catch (error) {
 	console.error(error instanceof Error ? error.message : String(error));

.claude/skills/verify/net.mjs

@@ -32,6 +32,7 @@ const instance = resolveInstanceOptions(
 );
 const CALL_TIMEOUT_MS = 15000;
 const FIXTURE_TRANSCRIPTION = "This text came from the fixture provider.";
+const FIXTURE_REPLY = "A second speaker answers.";
 const PATH_HOSTS = [FIXTURE_HOST, "api.openai.com", "api.deepgram.com"];
 
 /** @param {string} code */
@@ -75,6 +76,15 @@ const WATCH = `(() => {
 		}
 	};
 	const text = ${jsLiteral(FIXTURE_TRANSCRIPTION)};
+	// Deepgram's words: one every 0.4 s, the reply 12 s in from a second speaker.
+	const words = (sentence, speaker, from) => sentence.split(" ").map((word, index) => ({
+		word: word.toLowerCase().replace(/[^a-z]/g, ""),
+		punctuated_word: word,
+		start: from + index * 0.4,
+		end: from + index * 0.4 + 0.3,
+		speaker,
+	}));
+	const deepgram = { results: { channels: [{ alternatives: [{ words: [...words(text, 0, 0), ...words(${jsLiteral(FIXTURE_REPLY)}, 1, 12)] }] }] } };
 	const originalFetch = window.fetch;
 	window.fetch = async function (input, init) {
 		const url = typeof input === "string" ? input : input instanceof URL ? input.href : input.url;
@@ -88,12 +98,12 @@ const WATCH = `(() => {
 		if (channel === "request-url" && typeof request?.url === "string") {
 			const method = (request.method ?? "GET").toUpperCase();
 			const answer = request.url.startsWith("https://api.openai.com/")
-				? { text, segments: [{ speaker: "A", text, start: 0, end: 1 }] }
+				? { text }
 				: request.url.startsWith("https://api.deepgram.com/")
-					? { results: { utterances: [{ speaker: 0, transcript: text, start: 0, end: 1 }] } }
+					? deepgram
 					: null;
 			if (answer) {
-				entries.push({ via: "requestUrl", method, url: logged(request.url), answeredBy: "fixture provider", uploadBytes: request.body?.byteLength ?? 0 });
+				entries.push({ via: "requestUrl", method, url: logged(request.url), answeredBy: "fixture provider", contentType: request.contentType, uploadBytes: request.body?.byteLength ?? 0 });
 				const body = new TextEncoder().encode(JSON.stringify(answer)).buffer;
 				queueMicrotask(() => ipc.emit(replyChannel, {}, { status: 200, headers: { "content-type": "application/json" }, body }));
 				return;

.oxlintrc.json

@@ -32,6 +32,23 @@
 		]
 	},
 	"overrides": [
+		{
+			"files": ["src/**/*.ts", "tests/**/*.ts"],
+			"excludeFiles": ["src/services/transcription/**"],
+			"rules": {
+				"no-restricted-imports": [
+					"error",
+					{
+						"patterns": [
+							{
+								"group": ["**/services/transcription/**", "./transcription/**"],
+								"message": "Import from src/services/transcription, the folder's public surface."
+							}
+						]
+					}
+				]
+			}
+		},
 		{
 			"files": ["src/**/*.ts"],
 			"excludeFiles": ["src/**/*.test.ts", "src/**/*.spec.ts"],

AGENTS.md

@@ -12,6 +12,10 @@ in `src/main.ts`; public API code is under `src/API/`; parsing code is under
 are under `src/utility/`; Svelte UI lives under `src/ui/`; shared types live in
 `src/types/`.
 
+Audio transcription lives in `src/services/transcription/`. A provider is one
+file under `providers/` and one line in `registry.ts`. Code outside the folder
+imports only its `index.ts`, which `npm run lint` enforces.
+
 Tests are colocated with source files as `*.test.ts` where practical. Shared
 test mocks live in `tests/mocks/`. User-facing documentation lives in `docs/`
 and is built with Astro Starlight.
@@ -58,7 +62,7 @@ svelte-check's native API path passes when a tsconfig matches no files, which
 `check:a11y` in temporary workspaces and requires it to fail on that case and
 when the native compiler cannot start.
 
-After `npm run build`, `npm run size` enforces 330,000 raw bytes and 104,145 gzip bytes at level 9 for root `main.js` in the Test workflow's Build and sizes worker. `npm run size:data` reports a synthetic v2 stress library with 50 feeds, 2,000 history entries, and 200 saved episodes from frozen NPR metadata, including duplicated snapshots. Its 1.5 MB document goal is advisory. Pass an additive candidate JSON path to report optional-field overhead without changing existing data. In the isolated Obsidian instance, `npm run count:writes -- start`, `report`, and `stop` record writes to PodNotes' exact `data.json` path. Stop the recorder before stopping the instance, because recorder commands start a stopped instance; plugin reloads invalidate its counts.
+After `npm run build`, `npm run size` enforces 330,000 raw bytes and 104,290 gzip bytes at level 9 for root `main.js` in the Test workflow's Build and sizes worker. `npm run size:data` reports a synthetic v2 stress library with 50 feeds, 2,000 history entries, and 200 saved episodes from frozen NPR metadata, including duplicated snapshots. Its 1.5 MB document goal is advisory. Pass an additive candidate JSON path to report optional-field overhead without changing existing data. In the isolated Obsidian instance, `npm run count:writes -- start`, `report`, and `stop` record writes to PodNotes' exact `data.json` path. Stop the recorder before stopping the instance, because recorder commands start a stopped instance; plugin reloads invalidate its counts.
 
 Before opening a PR or cutting a release, run the CI-equivalent checks locally:
 

scripts/size-baseline.md

@@ -12,7 +12,7 @@ main.js raw: 307131 bytes; budget 330000 bytes; 22869 bytes remaining
 main.js gzip (level 9): 97161 bytes; budget 100000 bytes; 2839 bytes remaining
 ```
 
-The gzip budget is now 104,145 bytes. On 2026-10-14 it rose from 100,000 for
+The gzip budget is now 104,290 bytes. On 2026-10-14 it rose from 100,000 for
 the full player's Transcript panel, which the owner approved at about 1.9 KB
 and which measured 2,576 bytes. The coordinator set 103,000 to keep the
 panel's Find and Follow chip and to leave room for fixes in flight. The
@@ -43,6 +43,17 @@ main.js raw: 320267 bytes; budget 330000 bytes; 9733 bytes remaining
 main.js gzip (level 9): 104145 bytes; budget 104145 bytes; 0 bytes remaining
 ```
 
+On 2026-11-14 it rose from 104,145 to 104,290 for the transcription engine,
+which measured 104,060 on its own. On top of it, 170 bytes measure audio in other
+formats against each model's length cap, for at most 5 s and only while the job
+runs, so a long `.m4a` under the size cap is split instead of sent whole, and 60
+keep a signed enclosure URL from going to a provider. With them:
+
+```text
+main.js raw: 319405 bytes; budget 330000 bytes; 10595 bytes remaining
+main.js gzip (level 9): 104290 bytes; budget 104290 bytes; 0 bytes remaining
+```
+
 Gzip uses Node's `gzipSync` at level 9, without a filename or timestamp.
 Budgets use decimal bytes. The brief's 96,533-byte reference is unexplained.
 This report fixes the compressor instead of comparing compressed sizes across

scripts/size.mjs

@@ -10,7 +10,7 @@ try {
 	if (!bundle.length) throw new Error(`${path} is empty. Run npm run build first.`);
 	const measurements = [
 		{ name: "raw", bytes: bundle.length, budget: 330_000 },
-		{ name: "gzip (level 9)", bytes: gzipSync(bundle, { level: 9 }).length, budget: 104_145 },
+		{ name: "gzip (level 9)", bytes: gzipSync(bundle, { level: 9 }).length, budget: 104_290 },
 	];
 	const rows = measurements.map(({ name, bytes, budget }) => {
 		const overage = Math.max(0, bytes - budget);

scripts/size.test.ts

@@ -30,7 +30,7 @@ describe("bundle budget CLI", () => {
 		const result = check("gzip.js", randomBytes(110_000));
 		expect(result.status).toBe(1);
 		expect(result.stdout).toMatch(
-			/gzip \(level 9\): \d+ bytes; budget 104145 bytes; OVER by \d+ bytes/,
+			/gzip \(level 9\): \d+ bytes; budget 104290 bytes; OVER by \d+ bytes/,
 		);
 		expect(result.stdout).toContain("budget 330000 bytes; 220000 bytes remaining");
 	});

src/constants.ts

@@ -1,6 +1,5 @@
 import type { IPodNotesSettings } from "src/types/IPodNotesSettings";
 import type { Playlist } from "./types/Playlist";
-import { DEFAULT_SPEAKER_TEMPLATE } from "./services/diarization/segments";
 import { ownRecord } from "./utility/ownRecord";
 
 export const VIEW_TYPE = "podcast_player_view";
@@ -162,12 +161,10 @@ export const DEFAULT_SETTINGS: IPodNotesSettings = {
 	transcript: {
 		path: "transcripts/{{podcast}}/{{title}}.md",
 		template: "# {{title}}\n\nPodcast: {{podcast}}\nDate: {{date}}\n\n{{transcript}}",
-		// Diarization is off by default so existing behaviour (plain Whisper) is
-		// unchanged; enabling it routes audio to the chosen provider (#168).
 		diarization: {
 			enabled: false,
 			provider: "openai",
-			speakerTemplate: DEFAULT_SPEAKER_TEMPLATE,
+			speakerTemplate: "**{{speaker}}:** ",
 		},
 	},
 };