dsh-line-bot 0.9.7 → 0.10.1

package/lib/host.mjs

@@ -29,6 +29,7 @@ const stickersMod = await import(withVersion('./stickers.mjs'))
 const memeSourceMod = await import(withVersion('./meme-source.mjs'))
 const stickerImageMod = await import(withVersion('./sticker-image.mjs'))
 const visionMod = await import(withVersion('./vision.mjs'))
+const ttsMod = await import(withVersion('./tts.mjs'))
 const utilMod = await import(withVersion('./util.mjs'))
 const { createConfigStore, publicConfig, dataDir, dshHome } = configMod
 const { createPersonaClient } = personaMod
@@ -47,9 +48,10 @@ const { createStickerStore } = stickersMod
 const { createMemeSource } = memeSourceMod
 const { createStickerImages } = stickerImageMod
 const { createVision } = visionMod
+const { createTts } = ttsMod
 const { chunkText, clip, inQuietWindow, jsonFail, jsonOk, maskSecret, readJsonBody, safeArray, sleep, readJsonSafe } = utilMod
 
-export const VERSION = '0.9.7'
+export const VERSION = '0.10.1'
 
 // 環境變數提供的 Cloudflare API Token（方便用 shell 注入、不用把 token 寫進設定檔）。
 // 設定檔裡的 token 優先；兩者都沒有時才用環境變數。
@@ -401,7 +403,77 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
     const jokeText = jokes.contextText(() => store.get())
     // 群組不注入貼吧作息，避免在群組裡突然講起刷貼吧的事。
     const tieba = kind === 'user' ? tiebaContextText() : ''
-    return [modelContextText(), jokeText, tieba, balanceContextText(), extra, memoryContextText(ownerKey), memeContextText() || stickerContextText()].filter(Boolean).join('\n\n')
+    return [modelContextText(), voiceContextText(), jokeText, tieba, balanceContextText(), extra, memoryContextText(ownerKey), memeContextText() || stickerContextText()].filter(Boolean).join('\n\n')
+  }
+
+  const VOICE_MARKER = /\[\[\s*(?:voice|語音|语音|audio)\s*(?:[:：]\s*([^\]\n]{0,200}))?\s*\]\]/gi
+  let voiceCache = { at: 0, value: null }
+
+  function voiceStatusCached() {
+    if (!voiceCache.value || Date.now() - voiceCache.at > 10000) {
+      try {
+        voiceCache = { at: Date.now(), value: tts.status() }
+      } catch (error) {
+        voiceCache = { at: Date.now(), value: { ready: false, lastError: errorText(error) } }
+      }
+    }
+    return voiceCache.value
+  }
+
+  function voiceReady(cfg = store.get()) {
+    if (cfg.voice?.enabled !== true || cfg.voice?.consent !== true) return false
+    const st = voiceStatusCached()
+    return st?.ready === true && Boolean(st?.reference?.path)
+  }
+
+  function voiceContextText() {
+    const cfg = store.get()
+    if (cfg.voice?.enabled !== true || cfg.voice?.consent !== true) return ''
+    if (!voiceReady(cfg)) return ''
+    const max = Math.max(20, Math.min(300, Number(cfg.voice?.maxChars) || 80))
+    return [
+      '【語音】你可以用語音回覆。',
+      `想用時把 [[voice]] 放在回覆最後；如果要指定語氣，寫成 [[voice:用溫柔、帶點撒嬌的語氣，語速慢一點]]。`,
+      `語氣、情緒、語速、停頓這些「怎麼說」的指示只能放在標記裡；正文只放要念出來的字，不要把語氣描述、括號動作或舞台指示寫進正文（最多約 ${max} 字）。`,
+      '一次最多一則語音；不適合用語音（例如內容太長、需要圖片）就不要加標記。',
+    ].join('\n')
+  }
+
+  function splitVoiceMarker(text) {
+    let requested = false
+    let instruct = ''
+    const cleaned = String(text || '')
+      .replace(VOICE_MARKER, (_match, tone) => {
+        requested = true
+        if (tone && String(tone).trim()) instruct = String(tone).trim().slice(0, 200)
+        return ''
+      })
+      .trim()
+    return { text: cleaned, requested, instruct }
+  }
+
+  async function synthesizeVoiceMessage(text, instruct = '') {
+    const base = publicBaseSync()
+    if (!base) throw new Error('目前沒有公開網址，語音訊息送不出去')
+    const result = await tts.synthesize(text, { instructText: instruct })
+    const token = registerFile(result.m4a, 'audio/mp4')
+    if (!token) throw new Error('無法把語音檔存進媒體庫')
+    return {
+      message: { type: 'audio', originalContentUrl: `${base}/media/${token}`, duration: Math.max(1000, Math.round(result.durationMs || 0)) },
+      token,
+      provider: result.provider,
+      durationMs: result.durationMs,
+    }
+  }
+
+  async function replyMessages(userId, replyToken, messages) {
+    if (!Array.isArray(messages) || messages.length === 0) return
+    try {
+      await withToken((token) => lineApi.replyMessage(token, replyToken, messages))
+    } catch (error) {
+      logger.warn(`reply 失敗，改用 push：${errorText(error)}`)
+      await withToken((token) => lineApi.pushMessage(token, userId, messages))
+    }
   }
 
   async function refreshBalance() {
@@ -1426,6 +1498,25 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
     return token
   }
 
+  function registerFile(filePath, mime, ext = '') {
+    if (!filePath || !fs.existsSync(filePath)) return ''
+    const token = `m${randomBytes(6).toString('hex')}`
+    const suffix = ext || path.extname(String(filePath || '')).toLowerCase() || '.bin'
+    const filename = `${token}${suffix}`
+    const target = path.join(mediaDir, filename)
+    try {
+      fs.copyFileSync(filePath, target)
+    } catch (error) {
+      logger.warn(`複製媒體失敗：${error?.message || error}`)
+      return ''
+    }
+    const item = { file: filename, mime: mime || 'application/octet-stream', at: Date.now(), bytes: fs.statSync(target).size }
+    mediaTokens.set(token, item)
+    mediaIndex[token] = item
+    saveMediaIndex()
+    return token
+  }
+
   function getMedia(id) {
     const key = String(id || '')
     if (!key) return null
@@ -1804,10 +1895,28 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
         cwd: ensureUserWorkspace(conversationKey) || cfg.agent.cwd || undefined,
       })
       let replyText = String(result?.text || '').trim()
+      const voiceDirective = splitVoiceMarker(replyText)
+      replyText = voiceDirective.text
       const parsedReply = splitStickerMarkers(replyText)
-      if (!parsedReply.text && parsedReply.names.length === 0) replyText = cfg.line.emptyReply || '（她沒有說話）'
-      await replyOrPush(chatId, event.replyToken, replyText)
-      runtime.lastReply = { chatId, kind: sourceKind, at: Date.now(), text: clip(parsedReply.text || (parsedReply.names.length ? `[表情包] ${parsedReply.names.join('、')}` : replyText), 300) }
+      let voiceSent = null
+      if (voiceDirective.requested && voiceReady(cfg)) {
+        const spoken = parsedReply.text || ''
+        if (spoken) {
+          try {
+            voiceSent = await synthesizeVoiceMessage(spoken, voiceDirective.instruct)
+            logger.info(`語音回覆：${spoken.length} 字，${Math.round((voiceSent.durationMs || 0) / 1000)} 秒（${voiceSent.provider}；語氣：${voiceDirective.instruct || '預設'}）`)
+          } catch (error) {
+            logger.warn(`語音合成失敗，改用文字：${errorText(error)}`)
+          }
+        }
+      }
+      if (voiceSent) {
+        await replyMessages(chatId, event.replyToken, [voiceSent.message])
+      } else {
+        if (!parsedReply.text && parsedReply.names.length === 0) replyText = cfg.line.emptyReply || '（她沒有說話）'
+        await replyOrPush(chatId, event.replyToken, replyText)
+      }
+      runtime.lastReply = { chatId, kind: sourceKind, at: Date.now(), text: clip(voiceSent ? `[語音] ${parsedReply.text || replyText}` : parsedReply.text || (parsedReply.names.length ? `[表情包] ${parsedReply.names.join('、')}` : replyText), 300) }
       store.updateState((state) => {
         const map = state[bucketName] || {}
         const target = map[conversationKey]
@@ -1834,6 +1943,7 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
   const memes = createMemeSource({ ctx, logger })
   const stickerImages = createStickerImages({ ctx, logger, dir })
   const vision = createVision({ logger, cacheDir: path.join(dir, 'cache') })
+  const tts = createTts({ dir, logger, getConfig: () => store.get() })
   const mediaTokens = new Map()
   let imageSupportCache = { at: 0, key: '', value: false }
 
@@ -1949,7 +2059,9 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
   })
 
   async function sendPush(userId, text) {
-    const { messages } = await buildOutgoingMessages(text)
+    const voiceDirective = splitVoiceMarker(text)
+    if (voiceDirective.requested) logger.info('主動訊息中的語音標記已忽略，改用文字')
+    const { messages } = await buildOutgoingMessages(voiceDirective.text)
     if (messages.length === 0) return { sent: false, reason: 'empty' }
     const imageCount = messages.filter((item) => typeof item !== 'string' && item.type === 'image').length
     if (imageCount > 0) logger.info(`主動訊息含 ${imageCount} 張圖片`)
@@ -2388,6 +2500,18 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
         imageVision: cfg.line?.imageVision !== false,
         stickerVision: cfg.line?.stickerVision !== false,
       },
+      voice: {
+        ...voiceStatusCached(),
+        enabled: cfg.voice?.enabled === true,
+        consent: cfg.voice?.consent === true,
+        provider: cfg.voice?.provider || 'auto',
+        cloneMode: cfg.voice?.cloneMode || 'instruct',
+        instructText: cfg.voice?.instructText || '',
+        maxChars: Number(cfg.voice?.maxChars) || 80,
+        referencePath: cfg.voice?.referencePath || '',
+        ready: voiceReady(cfg),
+        publicBase: publicBaseSync(),
+      },
       memory: {
         ...memory.stats(),
         enabled: cfg.memory?.enabled !== false,
@@ -2676,6 +2800,53 @@ export async function create({ ctx, config: pluginConfig, logger: upstream }) {
         const built = await buildOutgoingMessages(String(payload.text))
         return { messages: built.messages, parsed: built.parsed }
       }
+      case 'voice-detect':
+        return tts.detect()
+      case 'voice-install': {
+        if (tts.status().installing) return { started: false, already: true }
+        tts.install()
+          .then((result) => {
+            if (result?.provider) store.patch({ voice: { provider: result.provider } }, { silent: true })
+            logger.info(`語音環境安裝完成（${result?.provider || 'unknown'}）`)
+            voiceCache = { at: 0, value: null }
+          })
+          .catch((error) => logger.error(`語音環境安裝失敗：${errorText(error)}`))
+        return { started: true }
+      }
+      case 'voice-preview':
+      case 'voice-test': {
+        const text = String(payload.text || '主人，這是語音測試，聽得到嗎？').trim()
+        if (!text) throw new Error('缺少文字')
+        const result = await tts.synthesize(text)
+        const token = registerFile(result.m4a, 'audio/mp4')
+        const base = publicBaseSync()
+        voiceCache = { at: 0, value: null }
+        return {
+          provider: result.provider,
+          durationMs: result.durationMs,
+          bytes: result.bytes,
+          chars: result.chars,
+          processingMs: result.processingMs,
+          url: base ? `${base}/media/${token}` : '',
+          localUrl: `/api/line-bot/media/${token}`,
+          token,
+        }
+      }
+      case 'voice-save-reference': {
+        const dataUrl = String(payload.dataUrl || '')
+        const match = /^data:([^;]+);base64,([\s\S]+)$/.exec(dataUrl)
+        if (!match) throw new Error('缺少音檔資料')
+        const buffer = Buffer.from(match[2], 'base64')
+        if (buffer.length > 20 * 1024 * 1024) throw new Error('音檔超過 20MB')
+        const saved = await tts.saveReferenceBuffer(buffer, String(payload.filename || 'reference'), String(payload.ext || '.wav'))
+        store.patch({ voice: { referencePath: saved.path } }, { silent: true })
+        voiceCache = { at: 0, value: null }
+        return { path: saved.path, seconds: saved.seconds }
+      }
+      case 'voice-reset-reference':
+        store.patch({ voice: { referencePath: '' } }, { silent: true })
+        voiceCache = { at: 0, value: null }
+        return { path: tts.referencePath() }
       case 'meme-send': {
         const userId = String(payload.userId || '').trim()
         if (!userId) throw new Error('缺少 userId')