Subtitle translation and parallel processing

package.json

@@ -1,6 +1,6 @@
 {
 	"name": "rabbit-encoder",
-	"version": "7.3.1",
+	"version": "8.0.0",
 	"type": "module",
 	"scripts": {
 		"dev": "bun run --watch src/index.ts",

public/api/client.ts

@@ -393,3 +393,22 @@ export async function fetchVsDefaultEntry(presetId: string): Promise<VsFilterEnt
 	const res = await authFetch(`${API}/api/vs-presets/${encodeURIComponent(presetId)}/default-entry`);
 	return res.json();
 }
+
+export async function testTranslateConnection(opts: {
+	provider: "ollama" | "deepseek";
+	url?: string;
+	model: string;
+	apiKey?: string;
+	target?: string;
+}): Promise<{ ok: boolean; error?: string; sample?: string; target?: string }> {
+	try {
+		const res = await authFetch(`${API}/api/translate/test`, {
+			method: "POST",
+			headers: { "Content-Type": "application/json" },
+			body: JSON.stringify(opts),
+		});
+		return await res.json();
+	} catch (err: any) {
+		return { ok: false, error: err?.message || "Request failed" };
+	}
+}

public/config/options.ts

@@ -11,7 +11,6 @@ import type {
 	EncoderSpeed,
 	StyleAppearance,
 	SubtitleProcessingMode,
-	SubtitleStyle,
 	VideoEncodeMode,
 } from "../types";
 import type { PipelinePreset } from "../ui/models";
@@ -87,6 +86,18 @@ export const PIPELINE_PRESET_HELP: Record<PipelinePreset, string> = {
 	custom: "Configure each pipeline stage individually below.",
 };
 
+export const TRANSLATE_PROVIDERS = ["ollama", "deepseek"] as const;
+export type TranslateProviderOption = (typeof TRANSLATE_PROVIDERS)[number];
+
+export const TRANSLATE_PROVIDER_LABELS: Record<TranslateProviderOption, string> = {
+	ollama: "Ollama (local)",
+	deepseek: "DeepSeek (cloud)",
+};
+export const DEEPSEEK_MODELS = ["deepseek-v4-flash", "deepseek-v4-pro"] as const;
+
+export const TRANSLATE_STRATEGIES = ["translategemma", "generic"];
+export const TRANSLATE_MODEL_OPTIONS = ["translategemma:4b", "translategemma:12b", "translategemma:27b"];
+
 export const DEFAULT_STYLE_APPEARANCE: StyleAppearance = {
 	fontSize: 80,
 	primaryColour: "&H00FFFFFF",

public/features/preview.ts

@@ -102,9 +102,8 @@ export function renderPreviewSummary(samples: PreviewSample[]): void {
 	const minProjected = Math.min(...projections);
 	const maxProjected = Math.max(...projections);
 
-	const totalEncodedBytes = samples.reduce((a, s) => a + (s.encodedSizeBytes || 0), 0);
 	const totalSampledSec = samples.reduce((a, s) => a + (s.windowSeconds || 0), 0);
-	const avgBitrateKbps = totalSampledSec > 0 ? (totalEncodedBytes * 8) / 1000 / totalSampledSec : 0;
+	const avgBitrateKbps = count > 0 ? samples.reduce((a, s) => a + (s.encodedBitrateKbps || 0), 0) / count : 0;
 
 	el.style.display = "";
 	el.innerHTML = `

public/features/settings-controls.ts

@@ -520,6 +520,37 @@ export function renderLanguageFilterInput(container: HTMLElement, value: string[
 	container.appendChild(hint);
 }
 
+export function renderTranslationLanguagesInput(container: HTMLElement, value: string[], onChange: (value: string[]) => void): void {
+	container.innerHTML = "";
+
+	const label = document.createElement("label");
+	label.textContent = "Target languages";
+	container.appendChild(label);
+
+	const input = document.createElement("input");
+	input.type = "text";
+	input.className = "lang-filter-input";
+	input.placeholder = "eng, slv";
+	input.value = (value || []).join(", ");
+
+	const hint = document.createElement("div");
+	hint.className = "lang-filter-hint";
+	hint.textContent =
+		"Comma-separated ISO codes. Every listed language should end up with a full subtitle: " +
+		"if one is missing, it's translated with the LLM from an existing full subtitle.";
+
+	input.oninput = () =>
+		onChange(
+			input.value
+				.split(",")
+				.map((s) => s.trim())
+				.filter((s) => s.length > 0),
+		);
+
+	container.appendChild(input);
+	container.appendChild(hint);
+}
+
 export function renderNlmeansParamsEditor(container: HTMLElement, params: NlmeansLevelParams, onChange: (value: NlmeansLevelParams) => void): void {
 	container.innerHTML = "";
 	const grid = document.createElement("div");
@@ -840,7 +871,7 @@ export function renderFontDropdown(container: HTMLElement, value: string, fonts:
 	container.appendChild(label);
 }
 
-/** Free-text control (used for ASS &HAABBGGRR colours). */
+/** Free-text control. */
 export function renderTextControl(container: HTMLElement, label: string, value: string, placeholder: string, onChange: (v: string) => void): void {
 	container.innerHTML = "";
 	const wrap = document.createElement("label");
@@ -857,3 +888,51 @@ export function renderTextControl(container: HTMLElement, label: string, value:
 	wrap.appendChild(input);
 	container.appendChild(wrap);
 }
+
+/** Free-password control. */
+export function renderPasswordControl(container: HTMLElement, label: string, value: string, placeholder: string, onChange: (v: string) => void): void {
+	container.innerHTML = "";
+	const wrap = document.createElement("label");
+	wrap.className = "toggle-label";
+	const span = document.createElement("span");
+	span.textContent = `${label}\u00A0`;
+	const input = document.createElement("input");
+	input.type = "password";
+	input.className = "lang-filter-input";
+	input.placeholder = placeholder;
+	input.value = value;
+	input.oninput = () => onChange(input.value);
+	wrap.appendChild(span);
+	wrap.appendChild(input);
+	container.appendChild(wrap);
+}
+
+export function renderSelectControl(
+	container: HTMLElement,
+	labelText: string,
+	options: readonly string[],
+	value: string,
+	onChange: (value: string) => void,
+): void {
+	container.innerHTML = "";
+	const label = document.createElement("label");
+	label.className = "toggle-label";
+
+	const span = document.createElement("span");
+	span.textContent = `${labelText}\u00A0`;
+
+	const select = document.createElement("select");
+	select.className = "select-input";
+	for (const opt of options) {
+		const o = document.createElement("option");
+		o.value = opt;
+		o.textContent = opt;
+		if (opt === value) o.selected = true;
+		select.appendChild(o);
+	}
+	select.onchange = () => onChange(select.value);
+
+	label.appendChild(span);
+	label.appendChild(select);
+	container.appendChild(label);
+}

public/features/settings-form.ts

@@ -5,6 +5,7 @@ import {
 	AUDIO_ENCODE_OPTIONS,
 	CROP_OPTIONS,
 	DEBAND_LEVELS,
+	DEEPSEEK_MODELS,
 	DEFAULT_AUTO_THRESHOLDS,
 	DEFAULT_GRADFUN_PARAMS,
 	DEFAULT_NLMEANS_PARAMS,
@@ -17,6 +18,10 @@ import {
 	SUBTITLE_FORMAT_PRIORITY_OPTIONS,
 	SUBTITLE_PROCESSING_OPTIONS,
 	SUBTITLE_SOURCE_PRIORITY_OPTIONS,
+	TRANSLATE_MODEL_OPTIONS,
+	TRANSLATE_PROVIDERS,
+	TRANSLATE_PROVIDER_LABELS,
+	TRANSLATE_STRATEGIES,
 	VIDEO_ENCODE_OPTIONS,
 } from "../config/options";
 import {
@@ -34,15 +39,20 @@ import {
 	renderLanguagePriorityInput,
 	renderNoPhaseInvToggle,
 	renderNumberControl,
+	renderPasswordControl,
 	renderRadioPills,
 	renderRemoveCommentaryAudioToggle,
+	renderSelectControl,
 	renderSkipBoostingToggle,
 	renderSubtitleConfidenceControl,
 	renderSubtitleLangDetectControl,
 	renderSubtitleStyleTargets,
+	renderTextControl,
+	renderTranslationLanguagesInput,
 	wireEncoderControls,
 } from "./settings-controls";
 import { byId } from "../shared/dom";
+import { testTranslateConnection } from "../api/client";
 
 export type SettingsFormPrefix = "default" | "job";
 
@@ -311,4 +321,114 @@ export function renderSettingsForm(prefix: SettingsFormPrefix, settings: JobSett
 	renderAudioLanguagesInput(el("audio-languages"), settings.audioLanguages || [], (v) => (settings.audioLanguages = v));
 	renderLanguageFilterInput(el("subtitle-languages"), settings.subtitleLanguages || [], (v) => (settings.subtitleLanguages = v));
 	renderBitrateInputs(el("bitrates"), settings.audioBitrates, (ch, val) => (settings.audioBitrates[ch] = val));
+
+	renderLabeledToggle(el("translate-enabled"), settings.translateSubtitles, "Translate missing languages", (v) => {
+		settings.translateSubtitles = v;
+	});
+
+	// Provider-specific settings, re-rendered when the provider pill changes.
+	const providerSettings = el("translate-provider-settings");
+	const renderProviderSettings = () => {
+		providerSettings.innerHTML = "";
+		const provider = settings.translateProvider ?? "ollama";
+		const group = (cls = "toggle-group") => {
+			const d = document.createElement("div");
+			d.className = cls;
+			providerSettings.appendChild(d);
+			return d;
+		};
+
+		if (provider === "ollama") {
+			renderTextControl(group(), "Ollama URL", settings.translateOllamaUrl, "http://localhost:11434", (v) => {
+				settings.translateOllamaUrl = v.trim();
+			});
+			renderTextControl(group(), "Model", settings.translateModel, "translategemma:12b", (v) => {
+				settings.translateModel = v.trim();
+			});
+			const hint = document.createElement("div");
+			hint.className = "lang-filter-hint";
+			hint.textContent = "Any Ollama model tag. TranslateGemma models use the dedicated translation prompt; other models use the generic format automatically.";
+			providerSettings.appendChild(hint);
+
+			renderNumberControl(group(), "Context window (num_ctx)", settings.translateNumCtx ?? 8192, { min: 2048, max: 131072, step: 1024 }, (v) => {
+				settings.translateNumCtx = v;
+			});
+		} else {
+			renderSelectControl(group(), "Model", DEEPSEEK_MODELS, settings.translateDeepseekModel ?? DEEPSEEK_MODELS[0], (v) => {
+				settings.translateDeepseekModel = v;
+			});
+			renderPasswordControl(group(), "API key", settings.translateApiKey ?? "", "sk-...", (v) => {
+				settings.translateApiKey = v.trim();
+			});
+			renderNumberControl(group(), "Max output tokens", settings.translateMaxTokens ?? 8192, { min: 1024, max: 32768, step: 512 }, (v) => {
+				settings.translateMaxTokens = v;
+			});
+		}
+	};
+
+	// Provider pills with human labels (same pattern as the encoder picker).
+	const provEl = el("translate-provider");
+	provEl.innerHTML = "";
+	for (const p of TRANSLATE_PROVIDERS) {
+		const pill = document.createElement("div");
+		pill.className = `radio-pill${p === (settings.translateProvider ?? "ollama") ? " selected" : ""}`;
+		pill.textContent = TRANSLATE_PROVIDER_LABELS[p];
+		pill.onclick = () => {
+			provEl.querySelectorAll(".radio-pill").forEach((x) => x.classList.remove("selected"));
+			pill.classList.add("selected");
+			settings.translateProvider = p;
+			renderProviderSettings();
+		};
+		provEl.appendChild(pill);
+	}
+	renderProviderSettings();
+
+	renderTranslationLanguagesInput(el("translate-targets"), settings.translateTargetLanguages, (langs) => {
+		settings.translateTargetLanguages = langs;
+	});
+
+	renderNumberControl(el("translate-batch"), "Dialogs per request", settings.translateBatchSize, { min: 1, max: 1000, step: 1 }, (v) => {
+		settings.translateBatchSize = v;
+	});
+
+	renderNumberControl(
+		el("translate-timeout"),
+		"Request timeout (s)",
+		Math.round((settings.translateTimeoutMs ?? 180_000) / 1000),
+		{ min: 10, max: 3600, step: 10 },
+		(v) => (settings.translateTimeoutMs = v * 1000),
+	);
+
+	renderNumberControl(el("translate-concurrency"), "Parallel requests", settings.translateConcurrency ?? 1, { min: 1, max: 16, step: 1 }, (v) => {
+		settings.translateConcurrency = v;
+	});
+
+	renderLabeledToggle(el("translate-signs"), settings.translateSignsSongs, "Also translate signs & songs", (v) => {
+		settings.translateSignsSongs = v;
+	});
+
+	const testBtn = el("translate-test") as HTMLButtonElement;
+	const testResult = el("translate-test-result");
+	testBtn.addEventListener("click", async () => {
+		testBtn.disabled = true;
+		testResult.textContent = "Testing…";
+		testResult.className = "test-result";
+		const provider = settings.translateProvider ?? "ollama";
+		const model = provider === "ollama" ? settings.translateModel : (settings.translateDeepseekModel ?? DEEPSEEK_MODELS[0]);
+		const r = await testTranslateConnection({
+			provider,
+			url: settings.translateOllamaUrl,
+			model,
+			apiKey: settings.translateApiKey,
+			target: settings.translateTargetLanguages[0],
+		});
+		testBtn.disabled = false;
+		if (r.ok) {
+			testResult.textContent = `✓ OK - sample (${r.target}): "${r.sample}"`;
+			testResult.classList.add("ok");
+		} else {
+			testResult.textContent = `✗ ${r.error}`;
+			testResult.classList.add("error");
+		}
+	});
 }

public/index.html

@@ -279,6 +279,31 @@ <h2>Default Settings</h2>
 										</div>
 									</div>
 								</details>
+								<details class="settings-subsection">
+									<summary>Subtitle translation</summary>
+									<div class="settings-subsection-body">
+										<div class="setting-group">
+											<div id="default-translate-enabled" class="toggle-group"></div>
+										</div>
+										<div class="setting-group">
+											<label>Provider</label>
+											<div class="radio-group" id="default-translate-provider"></div>
+											<div id="default-translate-provider-settings"></div>
+										</div>
+										<div class="setting-group">
+											<div id="default-translate-targets" class="toggle-group"></div>
+											<div id="default-translate-batch" class="toggle-group"></div>
+											<div id="default-translate-timeout" class="toggle-group"></div>
+											<div id="default-translate-concurrency" class="toggle-group"></div>
+											<div id="default-translate-signs" class="toggle-group"></div>
+										</div>
+										<div class="setting-group">
+											<button class="btn btn-ghost" id="default-translate-test">Test connection</button>
+											<br /><br />
+											<span id="default-translate-test-result"></span>
+										</div>
+									</div>
+								</details>
 								<div class="setting-group">
 									<label>Subtitle language priority</label>
 									<div id="default-subtitle-lang-priority"></div>
@@ -512,6 +537,31 @@ <h2 id="job-modal-title">Job Settings</h2>
 										</div>
 									</div>
 								</details>
+								<details class="settings-subsection">
+									<summary>Subtitle translation</summary>
+									<div class="settings-subsection-body">
+										<div class="setting-group">
+											<div id="job-translate-enabled" class="toggle-group"></div>
+										</div>
+										<div class="setting-group">
+											<label>Provider</label>
+											<div class="radio-group" id="job-translate-provider"></div>
+											<div id="job-translate-provider-settings"></div>
+										</div>
+										<div class="setting-group">
+											<div id="job-translate-targets" class="toggle-group"></div>
+											<div id="job-translate-batch" class="toggle-group"></div>
+											<div id="job-translate-timeout" class="toggle-group"></div>
+											<div id="job-translate-concurrency" class="toggle-group"></div>
+											<div id="job-translate-signs" class="toggle-group"></div>
+										</div>
+										<div class="setting-group">
+											<button class="btn btn-ghost" id="job-translate-test">Test connection</button>
+											<br /><br />
+											<span id="job-translate-test-result"></span>
+										</div>
+									</div>
+								</details>
 								<div class="setting-group">
 									<label>Subtitle language priority</label>
 									<div id="job-subtitle-lang-priority"></div>