@goodandready/dsh-tts 0.3.22 → 0.4.0

package/lib/providers.js

@@ -9,14 +9,13 @@ export const PROVIDER_KEYS = [
   'edge', 'piper', 'espeak', 'minimax',
 ]
 
-// Провайдеры, говорящие на языке OpenAI: тот же путь /audio/speech, отличаются
-// только адресом, моделью и именем ключа. Каждый адрес проверен запросом без
-// ключа — все отвечают 401, то есть путь существует и ждёт ключ.
+// OpenAI-compatible providers share /audio/speech; they differ by base URL,
+// model, and key name. Each URL was probed without a key and returned 401.
 export const OPENAI_COMPATIBLE = {
   siliconflow: { baseUrl: 'https://api.siliconflow.cn/v1', keyEnv: 'SILICONFLOW_API_KEY' },
   deepinfra: { baseUrl: 'https://api.deepinfra.com/v1/openai', keyEnv: 'DEEPINFRA_API_KEY' },
   fireworks: { baseUrl: 'https://api.fireworks.ai/inference/v1', keyEnv: 'FIREWORKS_API_KEY' },
-  // Пользовательский endpoint: адрес и имя ключа приходят из настроек.
+  // Custom endpoint: base URL and key name come from settings.
   custom: { baseUrl: '', keyEnv: 'CUSTOM_TTS_API_KEY' },
 }
 
@@ -44,7 +43,7 @@ export const DEFAULT_MODELS = {
 export const DEFAULT_VOICES = {
   kokoro: 'af_bella',
   f5: 'default',
-  // Голоса у MiMo названы по-китайски; «冰糖» — их же значение по умолчанию.
+  // MiMo voices use Chinese names; 冰糖 is their default.
   mimo: '冰糖',
   minimax: '',
   siliconflow: 'FunAudioLLM/CosyVoice2-0.5B:alex',
@@ -75,12 +74,32 @@ function asBuffer(body) {
 
 export function makeProviders(deps, req) {
   const { resolveKey, fetchImpl, cfg } = deps
-  const { text, lang, signal, models, voices } = req
+  const { text, lang, signal: parentSignal, models, voices } = req
+  const cloudTimeoutMs = Number(cfg.cloudTimeoutMs) > 0 ? Number(cfg.cloudTimeoutMs) : 10000
+
+  function withCloudTimeout() {
+    const controller = new AbortController()
+    let timedOut = false
+    const timer = setTimeout(() => {
+      timedOut = true
+      controller.abort(new Error(`Cloud TTS timeout after ${cloudTimeoutMs}ms`))
+    }, cloudTimeoutMs)
+    if (parentSignal) {
+      if (parentSignal.aborted) controller.abort(parentSignal.reason)
+      else parentSignal.addEventListener('abort', () => controller.abort(parentSignal.reason), { once: true })
+    }
+    return {
+      signal: controller.signal,
+      cleanup: () => clearTimeout(timer),
+    }
+  }
 
   async function openai() {
     const key = await resolveKey(cfg.openaiKeyEnv)
     if (!key) return { ok: false, provider: 'openai', reason: `no ${cfg.openaiKeyEnv}` }
-    const res = await fetchImpl((cfg.openaiBaseUrl || 'https://api.openai.com/v1') + '/audio/speech', {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl((cfg.openaiBaseUrl || 'https://api.openai.com/v1') + '/audio/speech', {
       method: 'POST',
       headers: { authorization: `Bearer ${key}`, 'content-type': 'application/json' },
       body: JSON.stringify({
@@ -94,10 +113,13 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`OpenAI HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'openai', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
-  // Одна реализация на всех, кто говорит на языке OpenAI: адрес и имя ключа
-  // берутся из таблицы, остальное совпадает буква в букву.
+    // One implementation for all OpenAI-compatible providers: base URL and key
+    // name come from the table; everything else is identical.
   function openaiCompatible(key) {
     const spec = OPENAI_COMPATIBLE[key]
     return async function speak() {
@@ -106,7 +128,9 @@ export function makeProviders(deps, req) {
       if (!token) return { ok: false, provider: key, reason: `no ${keyEnv}` }
       const base = cfg[key + 'BaseUrl'] || spec.baseUrl
       if (!base) return { ok: false, provider: key, reason: 'no base url configured' }
-      const res = await fetchImpl(base + '/audio/speech', {
+      const { signal, cleanup } = withCloudTimeout()
+      try {
+        const res = await fetchImpl(base + '/audio/speech', {
         method: 'POST',
         headers: { authorization: `Bearer ${token}`, 'content-type': 'application/json' },
         body: JSON.stringify({
@@ -120,17 +144,22 @@ export function makeProviders(deps, req) {
       if (!res.ok) throw new Error(`${key} HTTP ${res.status}`)
       const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
       return { ok: audio.length > 0, provider: key, audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+      } finally {
+        cleanup()
+      }
     }
   }
 
-  // Xiaomi MiMo. Синтез идёт через чат: текст кладётся сообщением, а голос и
-  // формат — отдельным полем; готовый звук приезжает в ответе base64.
+    // Xiaomi MiMo. Synthesis goes through chat: text as a message, voice/format
+    // as a separate field; audio returns base64 in the response.
   async function mimo() {
     const key = await resolveKey(cfg.mimoKeyEnv)
     if (!key) return { ok: false, provider: 'mimo', reason: `no ${cfg.mimoKeyEnv}` }
     const base = (cfg.mimoBaseUrl || 'https://api.xiaomimimo.com/v1').replace(/\/+$/, '')
     const format = cfg.mimoFormat === 'wav' ? 'wav' : 'mp3'
-    const res = await fetchImpl(base + '/chat/completions', {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl(base + '/chat/completions', {
       method: 'POST',
       headers: { authorization: `Bearer ${key}`, 'content-type': 'application/json', accept: 'application/json' },
       body: JSON.stringify({
@@ -160,9 +189,12 @@ export function makeProviders(deps, req) {
       mime: format === 'wav' ? 'audio/wav' : 'audio/mpeg',
       reason: audio.length ? '' : 'empty audio',
     }
+    } finally {
+      cleanup()
+    }
   }
 
-  // MiniMax — через их утилиту: своего протокола здесь нет намеренно.
+  // MiniMax uses their CLI on purpose; no custom protocol here.
   async function minimax() {
     try {
       const out = await minimaxSpeak(text, {
@@ -181,7 +213,9 @@ export function makeProviders(deps, req) {
     const key = await resolveKey(cfg.elevenlabsKeyEnv)
     if (!key) return { ok: false, provider: 'elevenlabs', reason: `no ${cfg.elevenlabsKeyEnv}` }
     const voice = pick(DEFAULT_VOICES, voices, 'elevenlabs')
-    const res = await fetchImpl(`https://api.elevenlabs.io/v1/text-to-speech/${encodeURIComponent(voice)}`, {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl(`https://api.elevenlabs.io/v1/text-to-speech/${encodeURIComponent(voice)}`, {
       method: 'POST',
       headers: { 'xi-api-key': key, 'content-type': 'application/json', accept: 'audio/mpeg' },
       body: JSON.stringify({
@@ -193,6 +227,9 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`ElevenLabs HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'elevenlabs', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function google() {
@@ -200,7 +237,9 @@ export function makeProviders(deps, req) {
     if (!key) return { ok: false, provider: 'google', reason: `no ${cfg.googleKeyEnv}` }
     const model = pick(DEFAULT_MODELS, models, 'google')
     const voiceName = pick(DEFAULT_VOICES, voices, 'google')
-    const res = await fetchImpl(
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl(
       `https://generativelanguage.googleapis.com/v1beta/models/${encodeURIComponent(model)}:generateContent?key=${encodeURIComponent(key)}`,
       {
         method: 'POST',
@@ -221,6 +260,9 @@ export function makeProviders(deps, req) {
     if (!b64) return { ok: false, provider: 'google', reason: 'no inline audio' }
     const audio = Buffer.from(b64, 'base64')
     return { ok: audio.length > 0, provider: 'google', audio, mime: 'audio/wav', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function azure() {
@@ -238,7 +280,9 @@ export function makeProviders(deps, req) {
     const ssml = `<speak version="1.0" xml:lang="${lang || 'en-US'}">`
       + `<voice xml:lang="${lang || 'en-US'}" name="${voice}">`
       + `${inner}</voice></speak>`
-    const res = await fetchImpl(`https://${region}.tts.speech.microsoft.com/cognitiveservices/v1`, {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl(`https://${region}.tts.speech.microsoft.com/cognitiveservices/v1`, {
       method: 'POST',
       headers: {
         'Ocp-Apim-Subscription-Key': key,
@@ -251,12 +295,17 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`Azure HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'azure', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function groq() {
     const key = await resolveKey(cfg.groqKeyEnv)
     if (!key) return { ok: false, provider: 'groq', reason: `no ${cfg.groqKeyEnv}` }
-    const res = await fetchImpl('https://api.groq.com/openai/v1/audio/speech', {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl('https://api.groq.com/openai/v1/audio/speech', {
       method: 'POST',
       headers: { authorization: `Bearer ${key}`, 'content-type': 'application/json' },
       body: JSON.stringify({
@@ -270,13 +319,18 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`Groq HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'groq', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function deepgram() {
     const key = await resolveKey(cfg.deepgramKeyEnv)
     if (!key) return { ok: false, provider: 'deepgram', reason: `no ${cfg.deepgramKeyEnv}` }
     const model = pick(DEFAULT_MODELS, models, 'deepgram')
-    const res = await fetchImpl(`https://api.deepgram.com/v1/speak?model=${encodeURIComponent(model)}`, {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl(`https://api.deepgram.com/v1/speak?model=${encodeURIComponent(model)}`, {
       method: 'POST',
       headers: { authorization: `Token ${key}`, 'content-type': 'application/json' },
       body: JSON.stringify({ text }),
@@ -285,12 +339,17 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`Deepgram HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'deepgram', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function openrouter() {
     const key = await resolveKey(cfg.openrouterKeyEnv)
     if (!key) return { ok: false, provider: 'openrouter', reason: `no ${cfg.openrouterKeyEnv}` }
-    const res = await fetchImpl('https://openrouter.ai/api/v1/audio/speech', {
+    const { signal, cleanup } = withCloudTimeout()
+    try {
+      const res = await fetchImpl('https://openrouter.ai/api/v1/audio/speech', {
       method: 'POST',
       headers: { authorization: `Bearer ${key}`, 'content-type': 'application/json' },
       body: JSON.stringify({
@@ -304,6 +363,9 @@ export function makeProviders(deps, req) {
     if (!res.ok) throw new Error(`OpenRouter HTTP ${res.status}`)
     const audio = asBuffer(Buffer.from(await res.arrayBuffer()))
     return { ok: audio.length > 0, provider: 'openrouter', audio, mime: 'audio/mpeg', reason: audio.length ? '' : 'empty audio' }
+    } finally {
+      cleanup()
+    }
   }
 
   async function edge() {
@@ -340,6 +402,16 @@ export function makeProviders(deps, req) {
     try {
       const modelPath = deps.modelManager ? deps.modelManager.getModelPath('kokoro') : (cfg.kokoroModelPath || '')
       const engine = new KokoroEngine({ modelPath })
+      if (!engine.isInstalled()) {
+        return { ok: false, provider: 'kokoro', reason: 'Kokoro ONNX model not installed. Use Edge, Piper, or eSpeak for offline TTS.' }
+      }
+      if (!engine.isRuntimeAvailable()) {
+        return {
+          ok: false,
+          provider: 'kokoro',
+          reason: 'Kokoro ONNX runtime is not bundled. Weights alone do not produce speech. Use Edge, Piper, or eSpeak for offline TTS.',
+        }
+      }
       const voice = pick(DEFAULT_VOICES, voices, 'kokoro') || 'af_bella'
       const audio = await engine.synthesizeWav(text, voice)
       return { ok: audio && audio.length > 0, provider: 'kokoro', audio, mime: 'audio/wav' }
@@ -350,13 +422,21 @@ export function makeProviders(deps, req) {
 
   async function f5() {
     try {
-      const modelPath = deps.modelManager ? deps.modelManager.getModelPath('f5') : (cfg.f5ModelPath || '')
+      const modelPath = deps.modelManager ? deps.modelManager.getModelPath('f5') : ''
       const engine = new F5Engine({ modelPath })
       const ok = await engine.ping()
-      if (!ok) return { ok: false, provider: 'f5', reason: 'F5 daemon unavailable' }
-      const kokoroFallback = new KokoroEngine({ modelPath })
-      const audio = await kokoroFallback.synthesizeWav(text, 'af_bella')
-      return { ok: true, provider: 'f5', audio, mime: 'audio/wav' }
+      if (!ok) {
+        return { ok: false, provider: 'f5', reason: 'F5 daemon unavailable. Use Edge, Piper, or eSpeak for offline TTS.' }
+      }
+      if (!engine.isRuntimeAvailable()) {
+        return {
+          ok: false,
+          provider: 'f5',
+          reason: 'F5-TTS synthesis is not bundled. Use Edge, Piper, or eSpeak for offline TTS.',
+        }
+      }
+      const audio = await engine.synthesize(text)
+      return { ok: !!(audio && audio.length), provider: 'f5', audio, mime: 'audio/wav' }
     } catch (e) {
       return { ok: false, provider: 'f5', reason: String(e && e.message || e) }
     }