dsh-vidi 0.5.2 → 0.6.0

package/lib/client.js

@@ -821,12 +821,14 @@ function Pick(props) {
 		const room = window.innerHeight - r.bottom - 8, above = r.top - 8;
 		const up = room < 220 && above > room;
 		const maxH = Math.max(120, Math.min(320, up ? above - 4 : room - 4));
-		const width = Math.min(Math.max(r.width, 180), window.innerWidth - 16);
-		const left = Math.min(Math.max(8, r.left), window.innerWidth - width - 8);
+		const minWidth = Math.min(Math.max(r.width, 180), window.innerWidth - 16);
+		const maxWidth = Math.max(minWidth, Math.min(380, window.innerWidth - 16));
+		const side = r.left + maxWidth > window.innerWidth - 8 ? { right: Math.max(8, window.innerWidth - r.right) } : { left: Math.max(8, r.left) };
 		setPlace({
 			...vars,
-			left,
-			width,
+			...side,
+			minWidth,
+			maxWidth,
 			maxHeight: maxH,
 			...up ? { bottom: window.innerHeight - r.top + 4 } : { top: r.bottom + 4 },
 			fontFamily: cs.fontFamily
@@ -34605,6 +34607,19 @@ function RestreamPanel(props) {
 }
 
 // src/client/Settings.tsx
+/** Who runs a speech engine (engines of one vendor share its key). */
+const vendorOf = (p) => p.vendor ?? p.id;
+/**
+* Subtitle models worth suggesting first: the fast, cheap tier of each family
+* (flash / mini / turbo / air / lite / haiku), newest first, and the one DSH
+* uses by default.
+*/
+function recommended(models, fallback) {
+	const ver = (n) => Number(/(\d+(?:\.\d+)?)/.exec(n)?.[1] ?? 0);
+	const picks = models.filter((m) => /flash|mini|turbo|\bair\b|-air|lite|haiku/i.test(`${m.model} ${m.name}`) && !/embed|tts|audio|image|vision-preview|realtime/i.test(m.model)).sort((a, b) => ver(b.model) - ver(a.model)).slice(0, 2).map((m) => m.model);
+	if (fallback && models.some((m) => m.model === fallback)) picks.unshift(fallback);
+	return new Set(picks);
+}
 const engineLabel = (t, id) => {
 	const v = t(`engine.${id}`);
 	return v === `engine.${id}` ? id : v;
@@ -34628,6 +34643,13 @@ function Settings(props) {
 	const [msg, setMsg] = (0, react.useState)(null);
 	const cur = providers.find((p) => p.id === engine);
 	const key = cur?.key ?? { set: false };
+	const vendorName = (v) => {
+		const x = t(`vendor.${v}`);
+		return x === `vendor.${v}` ? name(v) : x;
+	};
+	const vendors = [...new Map(providers.map((p) => [vendorOf(p), p])).values()];
+	const curVendor = cur ? vendorOf(cur) : "local";
+	const vendorEngines = providers.filter((p) => vendorOf(p) === curVendor);
 	(0, react.useEffect)(() => {
 		if (!open) return;
 		const close = (e) => {
@@ -34646,9 +34668,6 @@ function Settings(props) {
 		fetch("/live/api/models").then((r) => r.json()).then(setModels).catch(() => {});
 	}, [open, models]);
 	const modelKey = (m) => m ? `${m.provider}/${m.model}` : "";
-	const modelName = (m) => {
-		return (models?.models.find((x) => modelKey(x) === modelKey(m)))?.name ?? m?.model ?? "";
-	};
 	const chooseModel = (key) => {
 		const row = models?.models.find((x) => modelKey(x) === key);
 		const subsModel = row ? {
@@ -34661,8 +34680,14 @@ function Settings(props) {
 		});
 		post$1("settings", { subsModel });
 	};
-	const providersOf = [...new Set((models?.models ?? []).map((m) => m.providerName))];
-	const current = models?.chosen ?? models?.fallback;
+	const current = models?.chosen ?? models?.fallback ?? models?.current ?? void 0;
+	const providersOf = [...new Map([...current && !(models?.models ?? []).some((m) => m.provider === current.provider) ? [[current.provider, {
+		id: current.provider,
+		name: current.provider
+	}]] : [], ...(models?.models ?? []).map((m) => [m.provider, {
+		id: m.provider,
+		name: m.providerName
+	}])]).values()];
 	const sees = !!models?.models.find((x) => modelKey(x) === modelKey(current))?.image;
 	const saveKey = async (value) => {
 		if (!cur) return;
@@ -34702,19 +34727,35 @@ function Settings(props) {
 			className: "dshl-pop settings right",
 			children: [
 				/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
-					className: "srow",
+					className: "srow col",
 					children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 						title: t("set.engine.tip"),
 						children: t("set.engine")
-					}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
-						value: engine,
-						disabled,
-						onChange: onEngine,
-						options: providers.map((p) => ({
-							value: p.id,
-							label: name(p.id),
-							hint: p.ready ? void 0 : t(p.online ? "set.engine.noKey" : "set.engine.notInstalled")
-						}))
+					}), /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+						className: "duo",
+						children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
+							value: curVendor,
+							disabled,
+							onChange: (v) => {
+								const first = providers.find((p) => vendorOf(p) === v);
+								if (first) onEngine(first.id);
+							},
+							options: vendors.map((p) => ({
+								value: vendorOf(p),
+								label: vendorName(vendorOf(p)),
+								hint: p.ready ? void 0 : t(p.online ? "set.engine.noKey" : "set.engine.notInstalled"),
+								group: p.online ? t(p.region === "cn" ? "set.engine.cn" : "set.engine.intl") : void 0
+							}))
+						}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
+							value: engine,
+							disabled: disabled || vendorEngines.length < 2,
+							onChange: onEngine,
+							options: vendorEngines.map((p, i) => ({
+								value: p.id,
+								label: p.model ?? name(p.id),
+								hint: t(`engine.note.${p.id}`) !== `engine.note.${p.id}` ? t(`engine.note.${p.id}`) : i === 0 && vendorEngines.length > 1 ? t("set.recommended") : void 0
+							}))
+						})]
 					})]
 				}),
 				cur?.online && /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
@@ -34722,7 +34763,7 @@ function Settings(props) {
 					children: [
 						/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 							title: t(shared ? "set.key.tip.shared" : "set.key.tip"),
-							children: t("set.key").replace("{s}", name(cur.id))
+							children: t("set.key").replace("{s}", vendorName(curVendor))
 						}),
 						key.set && !draft && /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
 							className: "keyline",
@@ -34783,28 +34824,60 @@ function Settings(props) {
 				}),
 				/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 					className: "srow col",
-					children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-						className: "pair",
-						children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+					children: [
+						/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 							title: t("set.model.tip"),
 							children: t("set.model")
-						}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
-							value: models?.chosen ? modelKey(models.chosen) : "",
-							onChange: chooseModel,
-							options: [{
-								value: "",
-								label: models?.fallback ? t("set.model.default").replace("{m}", modelName(models.fallback)) : t("set.model.dsh")
-							}, ...providersOf.flatMap((pn) => models.models.filter((m) => m.providerName === pn).map((m) => ({
-								value: modelKey(m),
-								label: m.name,
-								hint: m.image ? t("set.model.image") : void 0,
-								group: pn
-							})))]
-						})]
-					}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-						className: "hint",
-						children: t(sees ? "set.model.sees" : "set.model.text")
-					})]
+						}),
+						/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+							className: "duo",
+							children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
+								value: current?.provider ?? "",
+								placeholder: "…",
+								onChange: (v) => {
+									const list = (models?.models ?? []).filter((m) => m.provider === v);
+									const rec = recommended(list, models?.fallback?.provider === v ? models.fallback.model : void 0);
+									const pick = list.find((m) => rec.has(m.model)) ?? list[0];
+									if (pick) chooseModel(modelKey(pick));
+								},
+								options: providersOf.map((p) => ({
+									value: p.id,
+									label: p.name
+								}))
+							}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)(Pick, {
+								value: modelKey(current),
+								placeholder: "…",
+								onChange: chooseModel,
+								options: (() => {
+									const listed = (models?.models ?? []).filter((m) => m.provider === current?.provider);
+									const list = current && !listed.some((m) => m.model === current.model) ? [{
+										provider: current.provider,
+										providerName: current.provider,
+										model: current.model,
+										name: current.model,
+										image: false
+									}, ...listed] : listed;
+									const rec = recommended(list, models?.fallback?.provider === current?.provider ? models?.fallback?.model : void 0);
+									const tag = (m) => [modelKey(m) === modelKey(models?.fallback) ? t("set.model.dshDefault") : "", m.image ? t("set.model.image") : ""].filter(Boolean).join(" · ") || void 0;
+									return [...list.filter((m) => rec.has(m.model)).map((m) => ({
+										value: modelKey(m),
+										label: m.name,
+										hint: tag(m),
+										group: t("set.recommended")
+									})), ...list.filter((m) => !rec.has(m.model)).map((m) => ({
+										value: modelKey(m),
+										label: m.name,
+										hint: tag(m),
+										group: rec.size ? t("set.allModels") : void 0
+									}))];
+								})()
+							})]
+						}),
+						/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+							className: "hint",
+							children: t(sees ? "set.model.sees" : "set.model.text")
+						})
+					]
 				}),
 				/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 					className: "srow col",
@@ -36848,8 +36921,31 @@ const en = {
 	"set.engine.notInstalled": "not installed",
 	"engine.local": "This computer (Whisper)",
 	"engine.openai": "OpenAI",
+	"engine.openai-translate": "OpenAI live translation",
+	"set.engine.intl": "Online",
+	"set.recommended": "Recommended",
+	"set.allModels": "All models",
+	"set.model.dshDefault": "DSH default",
+	"engine.note.openai-translate": "listens and translates",
+	"vendor.local": "This computer",
+	"vendor.openai": "OpenAI",
+	"vendor.gemini": "Google",
+	"vendor.deepgram": "Deepgram",
+	"vendor.elevenlabs": "ElevenLabs",
+	"vendor.mistral": "Mistral",
+	"vendor.groq": "Groq",
+	"vendor.dashscope": "Alibaba Cloud Bailian",
+	"vendor.volc": "Volcengine",
+	"vendor.siliconflow": "SiliconFlow",
+	"vendor.glm": "Zhipu",
+	"set.engine.cn": "Online · mainland China",
 	"engine.gemini": "Google Gemini",
+	"engine.deepgram": "Deepgram",
+	"engine.elevenlabs": "ElevenLabs",
+	"engine.mistral": "Mistral (Voxtral)",
 	"engine.groq": "Groq",
+	"engine.dashscope": "Alibaba Cloud (Qwen-ASR)",
+	"engine.volc": "Volcengine (Doubao)",
 	"engine.siliconflow": "SiliconFlow",
 	"engine.glm": "Zhipu GLM",
 	"set.local.missing": "Local recognition (whisper.cpp) isn’t installed on this computer. Pick an online service above and add its key.",
@@ -37184,8 +37280,31 @@ const zh = {
 	"set.engine.notInstalled": "未安装",
 	"engine.local": "本机（Whisper）",
 	"engine.openai": "OpenAI",
+	"engine.openai-translate": "OpenAI 实时翻译",
+	"set.engine.intl": "在线 · 海外",
+	"set.recommended": "推荐",
+	"set.allModels": "全部模型",
+	"set.model.dshDefault": "DSH 默认",
+	"engine.note.openai-translate": "边听边翻",
+	"vendor.local": "本机",
+	"vendor.openai": "OpenAI",
+	"vendor.gemini": "Google",
+	"vendor.deepgram": "Deepgram",
+	"vendor.elevenlabs": "ElevenLabs",
+	"vendor.mistral": "Mistral",
+	"vendor.groq": "Groq",
+	"vendor.dashscope": "阿里云百炼",
+	"vendor.volc": "火山引擎",
+	"vendor.siliconflow": "硅基流动",
+	"vendor.glm": "智谱",
+	"set.engine.cn": "在线 · 国内",
 	"engine.gemini": "Google Gemini",
+	"engine.deepgram": "Deepgram",
+	"engine.elevenlabs": "ElevenLabs",
+	"engine.mistral": "Mistral（Voxtral）",
 	"engine.groq": "Groq",
+	"engine.dashscope": "阿里云百炼（Qwen-ASR）",
+	"engine.volc": "火山引擎（豆包语音）",
 	"engine.siliconflow": "硅基流动",
 	"engine.glm": "智谱 GLM",
 	"set.local.missing": "这台电脑没装本地识别（whisper.cpp）。在上面换一个在线服务，填上它的密钥就能用。",
@@ -37795,7 +37914,9 @@ html[data-platform='darwin'] .dshl.page .dshl-top :is(button, input, label, a, .
 .dshl-pop.right { left: auto; right: 0; }
 .dshl-pop.up { top: auto; bottom: calc(100% + 6px); }
 .dshl .srow { display: flex; align-items: center; justify-content: space-between; gap: 12px; min-height: 46px; font-size: 13.5px; }
-.dshl-pop.settings { width: 320px; padding: 6px 12px; }
+.dshl-pop.settings { width: 380px; padding: 6px 12px; }
+.dshl .srow .duo { display: grid; grid-template-columns: minmax(0, 2fr) minmax(0, 3fr); gap: 8px; }
+.dshl .srow .duo > .dshl-select { width: 100%; }
 .dshl-pop.settings .srow + .srow { border-top: 1px solid var(--line-soft); }
 .dshl-pop.settings .srow > .dshl-select, .dshl-pop.settings .srow > .dshl-seg { width: 168px; }
 .dshl-pop.settings .srow > .dshl-seg button { flex: 1; padding: 0 6px; }

package/lib/index.mjs

@@ -2979,7 +2979,7 @@ var LiveJob = class LiveJob extends EventEmitter {
 // src/downloads.ts
 const run = promisify(execFile);
 /** ISO 639-2 codes for MP4 subtitle tracks. */
-const ISO3 = {
+const ISO3$1 = {
 	zh: "chi",
 	"zh-Hant": "chi",
 	en: "eng",
@@ -3125,7 +3125,7 @@ var Downloads = class {
 			const file = join(tmp, `${i}.srt`);
 			writeFileSync(file, exportCues(cues, t.langs, "srt", title));
 			inputs.push("-i", file);
-			meta.push(`-metadata:s:s:${i}`, `language=${ISO3[t.code] ?? ISO3[t.code.slice(0, 2)] ?? "und"}`, `-metadata:s:s:${i}`, `title=${t.name}`, `-disposition:s:${i}`, i === 0 ? "default" : "0");
+			meta.push(`-metadata:s:s:${i}`, `language=${ISO3$1[t.code] ?? ISO3$1[t.code.slice(0, 2)] ?? "und"}`, `-metadata:s:s:${i}`, `title=${t.name}`, `-disposition:s:${i}`, i === 0 ? "default" : "0");
 		});
 		try {
 			await run("ffmpeg", [
@@ -3300,7 +3300,12 @@ const KEY_REFS = {
 		"GLM_API_KEY",
 		"ZHIPUAI_API_KEY",
 		"ZAI_API_KEY"
-	]
+	],
+	deepgram: ["DEEPGRAM_API_KEY"],
+	elevenlabs: ["ELEVENLABS_API_KEY", "XI_API_KEY"],
+	mistral: ["MISTRAL_API_KEY"],
+	dashscope: ["DASHSCOPE_API_KEY", "BAILIAN_API_KEY"],
+	volc: ["VOLC_SPEECH_API_KEY", "VOLCENGINE_SPEECH_API_KEY"]
 };
 const KEY_NAMES = Object.keys(KEY_REFS);
 const mask = (k) => k.length > 10 ? `${k.slice(0, 3)}…${k.slice(-4)}` : "••••";
@@ -4071,39 +4076,107 @@ var OpenAIProvider = class {
 const SERVICES = [
 	{
 		id: "openai",
+		key: "openai",
 		label: "OpenAI",
 		keyUrl: "https://platform.openai.com/api-keys",
 		model: "gpt-live-transcribe",
+		region: "intl",
 		kind: "realtime"
 	},
+	{
+		id: "openai-translate",
+		key: "openai",
+		label: "OpenAI live translation",
+		keyUrl: "https://platform.openai.com/api-keys",
+		model: "gpt-realtime-translate",
+		region: "intl",
+		kind: "translate"
+	},
 	{
 		id: "gemini",
+		key: "gemini",
 		label: "Google Gemini",
 		keyUrl: "https://aistudio.google.com/apikey",
 		model: "auto",
+		modelLabel: "gemini-3.5-transcribe",
+		region: "intl",
 		kind: "gemini"
 	},
+	{
+		id: "deepgram",
+		key: "deepgram",
+		label: "Deepgram",
+		keyUrl: "https://console.deepgram.com/",
+		model: "nova-3",
+		region: "intl",
+		kind: "deepgram"
+	},
+	{
+		id: "elevenlabs",
+		key: "elevenlabs",
+		label: "ElevenLabs",
+		keyUrl: "https://elevenlabs.io/app/settings/api-keys",
+		model: "scribe_v2",
+		region: "intl",
+		kind: "elevenlabs"
+	},
+	{
+		id: "mistral",
+		key: "mistral",
+		label: "Mistral",
+		keyUrl: "https://console.mistral.ai/api-keys",
+		model: "voxtral-mini-latest",
+		region: "intl",
+		kind: "whisper-api",
+		base: "https://api.mistral.ai/v1"
+	},
 	{
 		id: "groq",
+		key: "groq",
 		label: "Groq",
 		keyUrl: "https://console.groq.com/keys",
 		model: "whisper-large-v3-turbo",
+		region: "intl",
 		kind: "whisper-api",
 		base: "https://api.groq.com/openai/v1"
 	},
+	{
+		id: "dashscope",
+		key: "dashscope",
+		label: "Alibaba Cloud Bailian",
+		keyUrl: "https://bailian.console.aliyun.com/?tab=model#/api-key",
+		model: "qwen3-asr-flash",
+		region: "cn",
+		kind: "dashscope",
+		base: "https://dashscope.aliyuncs.com/compatible-mode/v1"
+	},
+	{
+		id: "volc",
+		key: "volc",
+		label: "Volcengine Doubao",
+		keyUrl: "https://console.volcengine.com/speech/new/",
+		model: "volc.bigasr.auc_turbo",
+		modelLabel: "Seed-ASR 极速版",
+		region: "cn",
+		kind: "volc"
+	},
 	{
 		id: "siliconflow",
+		key: "siliconflow",
 		label: "SiliconFlow",
 		keyUrl: "https://cloud.siliconflow.cn/account/ak",
 		model: "FunAudioLLM/SenseVoiceSmall",
+		region: "cn",
 		kind: "whisper-api",
 		base: "https://api.siliconflow.cn/v1"
 	},
 	{
 		id: "glm",
+		key: "glm",
 		label: "Zhipu GLM",
 		keyUrl: "https://bigmodel.cn/usercenter/proj-mgmt/apikeys",
 		model: "glm-asr-2512",
+		region: "cn",
 		kind: "whisper-api",
 		base: "https://open.bigmodel.cn/api/paas/v4"
 	}
@@ -4123,6 +4196,8 @@ function speechServiceOf(id, baseURL) {
 	if (/api\.groq\.com|^ groq$/.test(k)) return "groq";
 	if (/siliconflow/.test(k)) return "siliconflow";
 	if (/bigmodel\.cn|(^|\.)z\.ai|^ (glm|zhipu|zhipuai|zai)$/.test(k)) return "glm";
+	if (/api\.mistral\.ai|^ mistral$/.test(k)) return "mistral";
+	if (/dashscope|maas\.aliyuncs\.com|^ (dashscope|bailian|aliyun|qwen)$/.test(k)) return "dashscope";
 }
 const TIMEOUT = 3e4;
 async function fail(r) {
@@ -4176,6 +4251,172 @@ function whisperApiEngine(svc, model, key) {
 		}
 	};
 }
+/** Glossary terms from a recogniser hint ("Glossary: a, b, c. …"), for services that take a word list. */
+const termsOf = (prompt) => (/Glossary: ([^\n]*?)\.(?:\s|$)/.exec(prompt)?.[1] ?? "").split(", ").map((x) => x.trim()).filter(Boolean);
+/** Three-letter language codes some services answer with, to the two-letter ones used here. */
+const ISO3 = {
+	eng: "en",
+	zho: "zh",
+	cmn: "zh",
+	yue: "yue",
+	jpn: "ja",
+	kor: "ko",
+	fra: "fr",
+	deu: "de",
+	spa: "es",
+	por: "pt",
+	rus: "ru",
+	ita: "it",
+	ara: "ar",
+	hin: "hi",
+	vie: "vi",
+	tha: "th",
+	ind: "id"
+};
+const lang2 = (l) => {
+	if (!l) return void 0;
+	const k = l.toLowerCase().split(/[-_]/)[0];
+	return ISO3[k] ?? ISO[k] ?? k;
+};
+/** Deepgram (Nova-3): one request per chunk, glossary as key terms, language detected per chunk. */
+function deepgramEngine(model, key) {
+	return {
+		name: "deepgram",
+		async transcribe(wav, lang, prompt) {
+			const apiKey = key();
+			if (!apiKey) throw new Error("no Deepgram key");
+			const q = new URLSearchParams({
+				model,
+				smart_format: "true",
+				punctuate: "true"
+			});
+			if (lang !== "auto") q.set("language", lang);
+			else q.set("detect_language", "true");
+			for (const t of termsOf(prompt ?? "").slice(0, 50)) q.append("keyterm", t);
+			const r = await fetch(`https://api.deepgram.com/v1/listen?${q}`, {
+				method: "POST",
+				headers: {
+					Authorization: `Token ${apiKey}`,
+					"content-type": "audio/wav"
+				},
+				body: new Uint8Array(wav),
+				signal: AbortSignal.timeout(TIMEOUT)
+			});
+			if (!r.ok) await fail(r);
+			const ch = (await r.json()).results?.channels?.[0];
+			return {
+				text: (ch?.alternatives?.[0]?.transcript ?? "").trim(),
+				lang: lang2(ch?.detected_language)
+			};
+		}
+	};
+}
+/** ElevenLabs Scribe: multipart upload per chunk; glossary as key terms. */
+function elevenLabsEngine(model, key) {
+	return {
+		name: "elevenlabs",
+		async transcribe(wav, lang, prompt) {
+			const apiKey = key();
+			if (!apiKey) throw new Error("no ElevenLabs key");
+			const form = new FormData();
+			form.append("file", new Blob([new Uint8Array(wav)], { type: "audio/wav" }), "chunk.wav");
+			form.append("model_id", model);
+			form.append("tag_audio_events", "false");
+			if (lang !== "auto") form.append("language_code", lang);
+			for (const t of termsOf(prompt ?? "").slice(0, 50)) form.append("keyterms", t);
+			const r = await fetch("https://api.elevenlabs.io/v1/speech-to-text", {
+				method: "POST",
+				headers: { "xi-api-key": apiKey },
+				body: form,
+				signal: AbortSignal.timeout(TIMEOUT)
+			});
+			if (!r.ok) await fail(r);
+			const j = await r.json();
+			return {
+				text: (j.text ?? "").replace(/\([^)]*\)/g, "").replace(/\s+/g, " ").trim(),
+				lang: lang2(j.language_code)
+			};
+		}
+	};
+}
+/** Alibaba Cloud Bailian (Qwen3-ASR-Flash), OpenAI-compatible chat call with the chunk inline. */
+function dashscopeEngine(base, model, key) {
+	return {
+		name: "dashscope",
+		async transcribe(wav, lang) {
+			const apiKey = key();
+			if (!apiKey) throw new Error("no Alibaba Cloud Bailian key");
+			const body = {
+				model,
+				stream: false,
+				messages: [{
+					role: "user",
+					content: [{
+						type: "input_audio",
+						input_audio: { data: `data:audio/wav;base64,${wav.toString("base64")}` }
+					}]
+				}],
+				asr_options: {
+					enable_itn: true,
+					...lang !== "auto" ? { language: lang } : {}
+				}
+			};
+			const r = await fetch(`${base}/chat/completions`, {
+				method: "POST",
+				headers: {
+					Authorization: `Bearer ${apiKey}`,
+					"content-type": "application/json"
+				},
+				body: JSON.stringify(body),
+				signal: AbortSignal.timeout(TIMEOUT)
+			});
+			if (!r.ok) await fail(r);
+			const m = (await r.json()).choices?.[0]?.message;
+			return {
+				text: (m?.content ?? "").replace(/\s+/g, " ").trim(),
+				lang: lang2(m?.annotations?.[0]?.language)
+			};
+		}
+	};
+}
+/** Volcengine Doubao speech (recording recognition, flash edition): one request per chunk, API key from the new speech console. */
+function volcEngine(resource, key) {
+	return {
+		name: "volc",
+		async transcribe(wav) {
+			const apiKey = key();
+			if (!apiKey) throw new Error("no Volcengine speech key");
+			const body = {
+				user: { uid: "dsh-vidi" },
+				audio: { data: wav.toString("base64") },
+				request: {
+					model_name: "bigmodel",
+					enable_itn: true,
+					enable_punc: true
+				}
+			};
+			const r = await fetch("https://openspeech.bytedance.com/api/v3/auc/bigmodel/recognize/flash", {
+				method: "POST",
+				headers: {
+					"X-Api-Key": apiKey,
+					"X-Api-Resource-Id": resource,
+					"X-Api-Request-Id": crypto.randomUUID(),
+					"X-Api-Sequence": "-1",
+					"content-type": "application/json"
+				},
+				body: JSON.stringify(body),
+				signal: AbortSignal.timeout(TIMEOUT)
+			});
+			const code = r.headers.get("X-Api-Status-Code");
+			if (code === "20000003") return { text: "" };
+			if (!r.ok || code && code !== "20000000") {
+				const msg = r.headers.get("X-Api-Message") ?? (await r.text().catch(() => "")).slice(0, 200);
+				throw new Error(`Volcengine ${code ?? r.status}: ${msg}`);
+			}
+			return { text: ((await r.json()).result?.text ?? "").trim() };
+		}
+	};
+}
 /** Newest Gemini Flash model the key can use (when the setting says "auto"). */
 async function pickGeminiModel(apiKey) {
 	const r = await fetch("https://generativelanguage.googleapis.com/v1beta/models?pageSize=200", {
@@ -4183,7 +4424,10 @@ async function pickGeminiModel(apiKey) {
 		signal: AbortSignal.timeout(TIMEOUT)
 	});
 	if (!r.ok) await fail(r);
-	const flash = ((await r.json()).models ?? []).filter((m) => m.supportedGenerationMethods?.includes("generateContent")).map((m) => m.name.replace(/^models\//, "")).filter((n) => /^gemini-\d+(\.\d+)?-flash(-preview)?$/.test(n));
+	const j = await r.json();
+	const transcribe = (j.models ?? []).map((m) => m.name.replace(/^models\//, "")).filter((n) => /^gemini-\d+(\.\d+)?-transcribe$/.test(n)).sort((a, b) => Number(/\d+(\.\d+)?/.exec(b)[0]) - Number(/\d+(\.\d+)?/.exec(a)[0]));
+	if (transcribe[0]) return transcribe[0];
+	const flash = (j.models ?? []).filter((m) => m.supportedGenerationMethods?.includes("generateContent")).map((m) => m.name.replace(/^models\//, "")).filter((n) => /^gemini-\d+(\.\d+)?-flash(-preview)?$/.test(n));
 	const version = (n) => Number(/^gemini-(\d+(?:\.\d+)?)/.exec(n)[1]);
 	flash.sort((a, b) => version(b) - version(a) || (a.includes("preview") ? 1 : 0) - (b.includes("preview") ? 1 : 0));
 	return flash[0] ?? "gemini-2.5-flash";
@@ -4210,6 +4454,7 @@ function geminiEngine(model, key, onUsage, log) {
 		async transcribe(wav, lang, prompt) {
 			const apiKey = key();
 			if (!apiKey) throw new Error("no Google Gemini key");
+			if (/-transcribe$/.test(chosen)) return transcribeModel(apiKey, chosen, wav, lang, prompt ?? "", onUsage);
 			const instruction = [
 				"Transcribe the speech in this audio exactly as spoken, in the language spoken (do not translate).",
 				lang !== "auto" ? `The speaker uses language code "${lang}".` : "",
@@ -4271,20 +4516,95 @@ function geminiEngine(model, key, onUsage, log) {
 		}
 	};
 }
+const BCP47 = {
+	zh: "zh-CN",
+	en: "en-US",
+	ja: "ja-JP",
+	ko: "ko-KR",
+	fr: "fr-FR",
+	de: "de-DE",
+	es: "es-ES",
+	pt: "pt-BR",
+	ru: "ru-RU",
+	it: "it-IT"
+};
+/** Gemini's dedicated transcription model (Interactions API): verbatim text, glossary as custom vocabulary. */
+async function transcribeModel(apiKey, model, wav, lang, prompt, onUsage) {
+	const vocab = termsOf(prompt).slice(0, 100);
+	const body = {
+		model: `models/${model}`,
+		input: [{
+			type: "audio",
+			data: wav.toString("base64"),
+			mime_type: "audio/wav"
+		}],
+		generation_config: { transcription_config: {
+			...lang !== "auto" ? { language_codes: [BCP47[lang] ?? lang] } : {},
+			...vocab.length ? { custom_vocabulary: vocab } : {},
+			mode: { type: "verbatim" }
+		} }
+	};
+	const r = await fetch("https://generativelanguage.googleapis.com/v1beta/interactions", {
+		method: "POST",
+		headers: {
+			"x-goog-api-key": apiKey,
+			"content-type": "application/json"
+		},
+		body: JSON.stringify(body),
+		signal: AbortSignal.timeout(TIMEOUT)
+	});
+	if (!r.ok) await fail(r);
+	const j = await r.json();
+	if (j.usage) onUsage(model, {
+		inputTokens: j.usage.total_input_tokens,
+		outputTokens: j.usage.total_output_tokens
+	});
+	const clean = (j.steps ?? []).filter((st) => st.type === "model_output").flatMap((st) => st.content ?? []).filter((c) => c.type === "text").map((c) => c.text ?? "").join("").replace(/\s+/g, " ").trim();
+	return {
+		text: clean,
+		lang: lang !== "auto" ? lang : scriptLang(clean)
+	};
+}
+/** The language a transcript is written in, when its script says so (kana, hangul, Han); else unknown. */
+function scriptLang(text) {
+	const n = (re) => (text.match(re) ?? []).length;
+	const han = n(/\p{Script=Han}/gu), kana = n(/[\p{Script=Hiragana}\p{Script=Katakana}]/gu), hangul = n(/\p{Script=Hangul}/gu);
+	const letters = n(/\p{L}/gu);
+	if (!letters) return void 0;
+	if (kana > letters * .1) return "ja";
+	if (hangul > letters * .3) return "ko";
+	if (han > letters * .3) return "zh";
+}
 /** Check a key before saving it: a cheap authenticated read on the service. */
 async function verifyKey(svc, key) {
 	try {
-		const url = svc.kind === "gemini" ? "https://generativelanguage.googleapis.com/v1beta/models?pageSize=1" : svc.id === "openai" ? "https://api.openai.com/v1/models" : svc.id === "glm" ? null : `${svc.base}/models`;
-		if (!url) return { ok: true };
-		const r = await fetch(url, {
-			headers: svc.kind === "gemini" ? { "x-goog-api-key": key } : { Authorization: `Bearer ${key}` },
+		const probe = svc.kind === "gemini" ? {
+			url: "https://generativelanguage.googleapis.com/v1beta/models?pageSize=1",
+			headers: { "x-goog-api-key": key }
+		} : svc.key === "openai" ? {
+			url: "https://api.openai.com/v1/models",
+			headers: { Authorization: `Bearer ${key}` }
+		} : svc.kind === "deepgram" ? {
+			url: "https://api.deepgram.com/v1/projects",
+			headers: { Authorization: `Token ${key}` }
+		} : svc.kind === "elevenlabs" ? {
+			url: "https://api.elevenlabs.io/v1/models",
+			headers: { "xi-api-key": key }
+		} : svc.id === "glm" || svc.kind === "volc" ? null : {
+			url: `${svc.base}/models`,
+			headers: { Authorization: `Bearer ${key}` }
+		};
+		if (!probe) return { ok: true };
+		const r = await fetch(probe.url, {
+			headers: probe.headers,
 			signal: AbortSignal.timeout(1e4)
 		});
 		if (r.ok) return { ok: true };
 		const body = await r.json().catch(() => ({}));
+		const detail = typeof body.detail === "object" ? body.detail?.message : body.detail;
 		return {
 			ok: false,
-			message: (typeof body.error === "object" ? body.error?.message : body.error) ?? body.message ?? `HTTP ${r.status}`
+			message: (typeof body.error === "object" ? body.error?.message : body.error) ?? body.message ?? detail ?? body.err_msg ?? `HTTP ${r.status}`
 		};
 	} catch (err) {
 		return {
@@ -4293,6 +4613,185 @@ async function verifyKey(svc, key) {
 		};
 	}
 }
+// src/providers/openai-translate.ts
+/** Output languages gpt-realtime-translate speaks (the input can be any of 70+). */
+const TRANSLATE_TARGETS = [
+	"es",
+	"pt",
+	"fr",
+	"ja",
+	"ru",
+	"zh",
+	"de",
+	"ko",
+	"hi",
+	"id",
+	"vi",
+	"it",
+	"en"
+];
+/**
+* OpenAI live interpretation (gpt-realtime-translate): audio streams in, the
+* translation and the source transcript stream back as text, so no separate
+* translation pass is needed. The service has no turns of its own, so the
+* shared pause detector decides where a subtitle ends; text that arrives
+* shortly after a pause still belongs to the line before it.
+*/
+var OpenAITranslateProvider = class {
+	opts;
+	name = "openai-translate";
+	translates = true;
+	ws;
+	chunker;
+	up = new Upsampler();
+	emit;
+	onError;
+	stopped = false;
+	reconnects = 0;
+	/** Text of the line being collected (source transcript and translation). */
+	src = "";
+	tr = "";
+	lineStart = 0;
+	seq = 0;
+	/** A finished stretch of audio whose text may still be arriving. */
+	closing;
+	constructor(opts) {
+		this.opts = opts;
+		if (!TRANSLATE_TARGETS.includes(opts.target)) throw new Error(`OpenAI live translation speaks only ${TRANSLATE_TARGETS.join(", ")}; choose one of these as the subtitle language, or another engine`);
+		this.chunker = new PauseChunker(opts.minSec, opts.maxSec, (c) => this.onChunk(c));
+	}
+	async start(emit, onError) {
+		this.emit = emit;
+		this.onError = onError;
+		await this.connect();
+	}
+	connect() {
+		return new Promise((resolve, reject) => {
+			const url = this.opts.url ?? `wss://api.openai.com/v1/realtime/translations?model=${encodeURIComponent(this.opts.model)}`;
+			const ws = new WebSocket(url, { headers: { Authorization: `Bearer ${this.opts.apiKey}` } });
+			this.ws = ws;
+			let opened = false;
+			ws.onopen = () => {
+				opened = true;
+				this.reconnects = 0;
+				this.send({
+					type: "session.update",
+					session: { audio: {
+						input: {
+							transcription: { model: "gpt-realtime-whisper" },
+							noise_reduction: { type: "far_field" }
+						},
+						output: { language: this.opts.target }
+					} }
+				});
+				resolve();
+			};
+			ws.onmessage = (ev) => {
+				try {
+					this.onEvent(JSON.parse(String(ev.data)));
+				} catch {}
+			};
+			ws.onerror = () => {
+				if (!opened) reject(/* @__PURE__ */ new Error("could not reach OpenAI live translation (check the API key and network)"));
+			};
+			ws.onclose = (ev) => {
+				if (!opened || this.stopped || this.ws !== ws) return;
+				this.opts.log?.(`openai-translate: connection closed (${ev.code} ${ev.reason || ""}), reconnecting`);
+				if (++this.reconnects > 5) {
+					this.onError?.(/* @__PURE__ */ new Error("OpenAI live translation keeps dropping"));
+					return;
+				}
+				setTimeout(() => {
+					if (!this.stopped) this.connect().catch((err) => this.onError?.(err));
+				}, 1e3 * this.reconnects);
+			};
+		});
+	}
+	send(event) {
+		if (this.ws?.readyState === WebSocket.OPEN) this.ws.send(JSON.stringify(event));
+	}
+	push(pcm) {
+		if (this.stopped) return;
+		this.send({
+			type: "session.input_audio_buffer.append",
+			audio: this.up.push(Buffer.from(pcm)).toString("base64")
+		});
+		this.chunker.push(pcm);
+	}
+	backlog() {
+		return this.closing ? this.chunker.now - this.closing.start : 0;
+	}
+	draft() {
+		if (!this.src.trim() && !this.tr.trim()) return;
+		this.emit?.({
+			key: `ot:${this.seq}`,
+			start: this.lineStart,
+			end: Math.max(this.chunker.now, this.lineStart + .5),
+			src: this.src.trim(),
+			tr: this.tr.trim() ? { [this.opts.target]: this.tr.trim() } : void 0,
+			final: false
+		});
+	}
+	/** A pause ended a stretch: wait a moment for its last words, then finish the line. */
+	onChunk({ start, end }) {
+		if (this.closing) this.finish();
+		if (!this.src.trim() && !this.tr.trim()) {
+			this.lineStart = end;
+			return;
+		}
+		this.lineStart = start;
+		this.closing = {
+			start,
+			end,
+			timer: setTimeout(() => this.finish(), 1500)
+		};
+	}
+	finish() {
+		const c = this.closing;
+		if (!c) return;
+		clearTimeout(c.timer);
+		this.closing = void 0;
+		const src = this.src.trim(), tr = this.tr.trim();
+		if (src || tr) this.emit?.({
+			key: `ot:${this.seq}`,
+			start: c.start,
+			end: Math.max(c.end, c.start + .5),
+			src: src || tr,
+			tr: tr ? { [this.opts.target]: tr } : void 0,
+			final: true
+		});
+		this.seq++;
+		this.src = "";
+		this.tr = "";
+		this.lineStart = c.end;
+	}
+	onEvent(ev) {
+		switch (ev.type) {
+			case "session.input_transcript.delta":
+				this.src += String(ev.delta ?? "");
+				this.draft();
+				break;
+			case "session.output_transcript.delta":
+				this.tr += String(ev.delta ?? "");
+				this.draft();
+				break;
+			case "error": {
+				const err = ev.error ?? {};
+				this.onError?.(/* @__PURE__ */ new Error(`OpenAI: ${err.message ?? "error"}`));
+				break;
+			}
+		}
+	}
+	async stop() {
+		this.chunker.flush(.5);
+		await new Promise((r) => setTimeout(r, 1500));
+		this.finish();
+		this.stopped = true;
+		try {
+			this.ws?.close();
+		} catch {}
+	}
+};
 // src/tools.ts
 const CLOCK = (s) => {
 	const h = Math.floor(s / 3600), m = Math.floor(s / 60) % 60, ss = Math.floor(s % 60);
@@ -4831,7 +5330,7 @@ function apply(ctx, config) {
 		].includes(x.name)).every((x) => x.ok);
 	});
 	/** Recognition to use when nothing was chosen: this computer if it can, else the first online service with a key. */
-	const defaultEngine = () => localReady ? "local" : SERVICES.find((s) => keys.get(s.id))?.id ?? "local";
+	const defaultEngine = () => localReady ? "local" : SERVICES.find((s) => keys.get(s.key))?.id ?? "local";
 	const settingsFile = join(config.dataDir, "settings.json");
 	const prefs = () => {
 		try {
@@ -4865,7 +5364,7 @@ function apply(ctx, config) {
 	function makeProvider(opts, glossary, usage, job) {
 		const svc = service(opts.provider);
 		if (svc) {
-			if (!keys.get(svc.id)) throw new Error(`no ${svc.label} key yet: add one in the video page settings or on the DSH Models page`);
+			if (!keys.get(svc.key)) throw new Error(`no ${svc.label} key yet: add one in the video page settings or on the DSH Models page`);
 			if (svc.kind === "realtime") return new OpenAIProvider({
 				apiKey: keys.get("openai"),
 				model: config.openaiModel,
@@ -4877,11 +5376,24 @@ function apply(ctx, config) {
 				log,
 				onUsage: (u) => usage.add("openai", config.openaiModel, u)
 			});
-			const key = () => keys.get(svc.id);
+			if (svc.kind === "translate") {
+				const target = opts.tgtLangs[0];
+				if (!target) throw new Error("OpenAI live translation needs a subtitle language to translate into");
+				return new OpenAITranslateProvider({
+					apiKey: keys.get("openai"),
+					model: svc.model,
+					target,
+					minSec: 2,
+					maxSec: 8,
+					log
+				});
+			}
+			const key = () => keys.get(svc.key);
+			const base = keys.bases.get(svc.key) ?? svc.base;
 			return new LocalProvider({
-				engine: svc.kind === "gemini" ? geminiEngine(config.geminiModel, key, (model, u) => usage.add("gemini", model, u), log) : whisperApiEngine({
+				engine: svc.kind === "gemini" ? geminiEngine(config.geminiModel, key, (model, u) => usage.add("gemini", model, u), log) : svc.kind === "deepgram" ? deepgramEngine(svc.model, key) : svc.kind === "elevenlabs" ? elevenLabsEngine(svc.model, key) : svc.kind === "dashscope" ? dashscopeEngine(base, svc.model, key) : svc.kind === "volc" ? volcEngine(svc.model, key) : whisperApiEngine({
 					...svc,
-					base: keys.bases.get(svc.id) ?? svc.base
+					base
 				}, svc.model, key),
 				lang: opts.srcLang === "auto" ? "auto" : opts.srcLang,
 				minSec: 3,
@@ -5052,14 +5564,19 @@ function apply(ctx, config) {
 					dataDir: config.dataDir,
 					providers: [{
 						id: "local",
+						vendor: "local",
+						model: "whisper-large-v3-turbo",
 						online: false,
 						ready: localReady
 					}, ...SERVICES.map((s) => ({
 						id: s.id,
+						vendor: s.key,
+						model: s.modelLabel ?? s.model,
 						label: s.label,
 						online: true,
-						ready: keys.info(s.id).set,
-						key: keys.info(s.id),
+						region: s.region,
+						ready: keys.info(s.key).set,
+						key: keys.info(s.key),
 						keyUrl: s.keyUrl
 					}))],
 					defaultEngine: defaultEngine(),
@@ -5076,10 +5593,10 @@ function apply(ctx, config) {
 						const check = await verifyKey(svc, key);
 						if (!check.ok) return send(res, 400, { error: check.message ?? "key rejected" });
 					}
-					await keys.save(svc.id, key);
-					return send(res, 200, { key: keys.info(svc.id) });
+					await keys.save(svc.key, key);
+					return send(res, 200, { key: keys.info(svc.key) });
 				}
-				const key = keys.get(svc.id);
+				const key = keys.get(svc.key);
 				if (!key) return send(res, 400, { error: "no key" });
 				const check = await verifyKey(svc, key);
 				return send(res, check.ok ? 200 : 400, check.ok ? { ok: true } : { error: check.message });
@@ -5196,7 +5713,7 @@ function apply(ctx, config) {
 					const svc = opts.provider ? service(opts.provider) : void 0;
 					if (svc) {
 						await keys.refresh();
-						if (!keys.get(svc.id)) return send(res, 400, { error: `no ${svc.label} key` });
+						if (!keys.get(svc.key)) return send(res, 400, { error: `no ${svc.label} key` });
 					}
 					savePrefs({ last: {
 						...lastOptions(),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-vidi",
-  "version": "0.5.2",
+  "version": "0.6.0",
   "description": "A video tool for DeepSeek Harness: watch live streams, videos and podcasts with real-time subtitles in any language, then clip and download",
   "type": "module",
   "license": "PolyForm-Noncommercial-1.0.0",