@xmanrui/dsh-im 4.28.1 → 4.29.0

package/src/channels/feishu/bridge.mjs

@@ -628,6 +628,14 @@ export class FeishuHarnessBridge {
   #logger;
   #signal;
   #botId;
+  /** Optional voice capability (语音输入/语音回复); null disables all voice paths. */
+  #voice = null;
+  /**
+   * Pending voice turns: session key → { replyTo }, consumed when the full
+   * answer lands. Keyed by the same session key as the message queue so
+   * concurrent topics of one group chat never share a voice slot.
+   */
+  #voiceTurns = new Map();
   #appId;
   #botOpenId;
   #groupResponseMode;
@@ -710,6 +718,7 @@ export class FeishuHarnessBridge {
     interactionCards = true,
     sessionSyncTargetsFor = null,
     logger = console,
+    voice = null,
     signal,
   }) {
     if (!client || !harness || !state || !status) {
@@ -736,6 +745,13 @@ export class FeishuHarnessBridge {
         || typeof stepPushClock.delay !== 'function')) {
       throw new TypeError('Feishu step push clock requires now and delay functions');
     }
+    if (voice !== null && voice !== undefined
+      && (typeof voice !== 'object' || Array.isArray(voice)
+        || (voice.enabled !== false
+          && (typeof voice.transcribeIncoming !== 'function'
+            || typeof voice.synthesize !== 'function')))) {
+      throw new TypeError('Feishu voice capability requires transcribeIncoming and synthesize functions');
+    }
     this.#client = client;
     this.#channel = channel;
     this.#harness = harness;
@@ -752,6 +768,7 @@ export class FeishuHarnessBridge {
     this.#stepPush = stepPush === true;
     this.#stepPushMode = normalizeFeishuStepPushMode(stepPushMode);
     this.#stepPushClock = stepPushClock ?? DEFAULT_STEP_PUSH_CLOCK;
+    this.#voice = voice ?? null;
     this.#repair = repair ?? null;
     this.#repairPollIntervalMs = repairPollIntervalMs;
     this.#repairLinkWaitMs = repairLinkWaitMs;
@@ -845,6 +862,10 @@ export class FeishuHarnessBridge {
     this.#stepPushMode = normalizeFeishuStepPushMode(value);
   }
 
+  setVoice(voice) {
+    this.#voice = voice ?? null;
+  }
+
   get stepPushMode() {
     return this.#stepPushMode;
   }
@@ -1494,10 +1515,57 @@ export class FeishuHarnessBridge {
     }
 
     const message = extractInboundMessage(event, this.#client);
-    const text = message.content;
+    let text = message.content;
+    let voiceTurnReplyTo = null;
+    // 语音输入(渠道能力):音频消息先转写成文字并写回 message.content——
+    // 下方命令识别与 #answerWithStream 都从 content 取文本。转写失败(或返回
+    // 空文本)时保持原语义,走"仅支持文字、图片和文件"的明确降级提示。
+    if (event.message.message_type === 'audio' && this.#voice?.enabled) {
+      try {
+        const transcript = await this.#voice.transcribeIncoming(event, this.#client);
+        if (transcript) {
+          text = transcript;
+          message.content = transcript;
+          voiceTurnReplyTo = event.message.message_id;
+        }
+      } catch (error) {
+        this.#logger.warn?.('[dsh-feishu-voice] 语音转写失败:', error?.message ?? String(error));
+      }
+    }
     const hasImages = hasInboundImages(message);
     const hasFiles = hasInboundFiles(message);
     const hasReply = hasReplyReference(message);
+    // 语音转写后的命令权限复检:accept() 的访问判定发生在转写前,音频消息
+    // 没有命令文本,按普通消息放行;转写得到的文本可能包含命令,须用与
+    // accept() 相同的命令识别与权限规则再校验一次,防止"允许聊天、禁止命令"
+    // 的用户经语音入口绕过 canExecuteCommands——文字与语音入口行为一致。
+    // 仅"命令不允许"给用户提示(与文字路径同文案),其余拒绝原因静默丢弃。
+    if (voiceTurnReplyTo) {
+      const transcribedAccess = evaluateInboundAccess(this.#accessPolicy, {
+        conversationType: event.message.chat_type === 'p2p' ? 'direct'
+          : event.message.chat_type === 'group' ? 'group' : null,
+        senderIds: senderOpenId(event),
+        text,
+        hasImages,
+        hasFiles,
+        isCommand: isSharedLocalCommand(text, { hasImages, hasFiles })
+          || isFeishuLocalCommand(text, { hasImages, hasFiles })
+          || (!hasImages && !hasFiles && NUMBER_REPLY.test(text) && this.#menus.has(key)),
+      });
+      if (!transcribedAccess.allowed) {
+        if (transcribedAccess.reason === 'command-not-allowed') {
+          await this.#send(
+            event.message.chat_id,
+            t(COMMAND_PERMISSION_DENIED_MESSAGE),
+            { replyTo: event.message.message_id },
+          );
+          this.#status.messagesReplied += 1;
+          this.#status.lastReplyAt = new Date().toISOString();
+        }
+        this.#status.lastError = null;
+        return;
+      }
+    }
     // 命令识别对 text 与纯文本 post 一视同仁：post 富文本若仅含单个
     // 文本段落（如复制粘贴的 /new），同样按命令处理；带图片/文件不认。
     // accept() 侧已用 nonEmptyString(content) 判定，两侧保持一致。
@@ -1637,6 +1705,15 @@ export class FeishuHarnessBridge {
     }
 
     this.#logger.info?.(`[dsh-feishu] processing ${event.message.chat_type} message ${messageId}`);
+    // 语音回合状态在进入正常问答路径前才登记:上方命令分支(/help、菜单、
+    // 会话列表等)与空内容分支均提前返回,提前登记会让状态遗留到下一回合,
+    // 导致后续普通文字消息的答案被合成为音频并回复到旧的语音消息。
+    // 本路径的 finally 统一清理回合状态。状态以会话 key 登记——与消息队列
+    // 同粒度:同群不同话题的 key 不同、可并行处理,共用 chat_id 槽位会让
+    // 后登记的回合覆盖先登记的,先完成的回合合成到后一条语音消息上。
+    if (voiceTurnReplyTo) {
+      this.#voiceTurns.set(key, { replyTo: voiceTurnReplyTo });
+    }
     const batchSubmission = event.batchSubmission ?? null;
     let batchAskCompleted = false;
     try {
@@ -1682,6 +1759,7 @@ export class FeishuHarnessBridge {
     } finally {
       await this.#cancelPendingInteraction(key);
       await this.#approvals.closeRoute(key);
+      this.#voiceTurns.delete(key);
     }
   }
 
@@ -2839,7 +2917,7 @@ export class FeishuHarnessBridge {
   }
 
   /**
-   * issue #162：已答状态卡的原地替换。patch 失败仅记录警告并明确降级——
+   * 已处理交互卡的原地替换。patch 失败仅记录警告并明确降级——
    * 回执缺失不应影响答案提交，也不得像 #sendCard 那样回退成发送新卡。
    */
   async #patchCardMessage(chatId, messageId, cardJson) {
@@ -2854,7 +2932,7 @@ export class FeishuHarnessBridge {
       }
       return messageId;
     } catch (error) {
-      this.#logger.warn?.('[dsh-feishu] answered-state card patch failed:', error?.message ?? error);
+      this.#logger.warn?.('[dsh-feishu] resolved interaction card patch failed:', error?.message ?? error);
       return null;
     }
   }
@@ -4142,6 +4220,53 @@ export class FeishuHarnessBridge {
     });
   }
 
+  // 语音回复(渠道能力):同回合转写过语音消息的会话,在完整答案投递后追加
+  // 音频回复。回合按会话 key 登记并消费——同群不同话题各自只消费自己的
+  // 回合;登记随消息处理结束(finally)清除,命令类回复(菜单、会话列表等)
+  // 与延迟交付不触发语音;此处失败只告警,不影响文字投递。
+  async #maybeSendVoiceReply(key, chatId, answer, replyTo = null) {
+    const voiceTurn = this.#voiceTurns.get(key);
+    if (!voiceTurn) return;
+    this.#voiceTurns.delete(key);
+    try {
+      await this.#sendVoiceAnswer(chatId, answer, voiceTurn.replyTo ?? replyTo);
+    } catch (error) {
+      this.#logger.warn?.('[dsh-feishu-voice] 语音回复失败:', error?.message ?? String(error));
+    }
+  }
+
+  // 文本 → qwen TTS → opus 上传 → 音频消息;优先回复原语音消息,
+  // reply 不可用时明确降级为普通音频消息。
+  async #sendVoiceAnswer(chatId, text, replyTo = null) {
+    const opus = await this.#voice.synthesize(text);
+    if (!opus) return null;
+    const upload = await this.#client.im.v1.file.create({
+      data: { file_type: 'opus', file_name: 'dsh-im-voice-reply.opus', file: opus },
+    });
+    const fileKey = upload?.file_key ?? upload?.data?.file_key;
+    if (!fileKey) throw new Error('音频上传无 file_key');
+    const content = JSON.stringify({ file_key: fileKey });
+    if (replyTo) {
+      try {
+        const response = await this.#client.im.v1.message.reply({
+          path: { message_id: replyTo },
+          data: { msg_type: 'audio', content },
+        });
+        if (!response?.code || response.code === 0) return response?.data?.message_id ?? null;
+      } catch (error) {
+        this.#logger.warn?.('[dsh-feishu-voice] 音频回复失败,改发普通消息:', error?.message ?? String(error));
+      }
+    }
+    const response = await this.#client.im.v1.message.create({
+      params: { receive_id_type: 'chat_id' },
+      data: { receive_id: chatId, msg_type: 'audio', content },
+    });
+    if (response?.code && response.code !== 0) {
+      throw new Error(`Feishu audio send failed: ${response.msg || response.code}`);
+    }
+    return nonEmptyString(response?.data?.message_id);
+  }
+
   async #deliverArtifacts(chatId, replyTo, artifacts = [], baseReceipt) {
     const delivery = await deliverOutboundArtifacts({
       artifacts,
@@ -5168,6 +5293,9 @@ export class FeishuHarnessBridge {
           providerMessageIds: seal.cardIds,
         });
         this.#status.streamResponses = (this.#status.streamResponses ?? 0) + 1;
+        // 流式卡封存成功也是完整答案投递:文本进了卡片,语音回合同样要
+        // 在此消费并合成音频回复——否则语音消息只得到卡片、没有声音。
+        await this.#maybeSendVoiceReply(key, chatId, deliveryText, messageId);
         const delivery = await this.#deliverArtifacts(
           chatId,
           messageId,
@@ -5217,6 +5345,7 @@ export class FeishuHarnessBridge {
         providerMessageIds,
       });
       this.#status.streamResponses = (this.#status.streamResponses ?? 0) + 1;
+      await this.#maybeSendVoiceReply(key, chatId, deliveryText, messageId);
     } catch (error) {
       this.#status.streamErrors = (this.#status.streamErrors ?? 0) + 1;
       this.#logger.warn?.('[dsh-feishu] step push post failed; sending final text:', error.message);
@@ -5267,8 +5396,10 @@ export class FeishuHarnessBridge {
         setLastMessageFailure(this.#status, textSendError);
       }
       this.#status.streamFallbacks = (this.#status.streamFallbacks ?? 0) + 1;
+      await this.#maybeSendVoiceReply(key, chatId, deliveryText, messageId);
       return { ...delivery, textDeliveryErrors: textSendError ? 1 : 0 };
     }
+    await this.#maybeSendVoiceReply(key, chatId, deliveryText, messageId);
     const delivery = await this.#deliverArtifacts(
       chatId,
       messageId,
@@ -5336,6 +5467,7 @@ export class FeishuHarnessBridge {
           error,
         );
       }
+      await this.#maybeSendVoiceReply(key, chatId, answerTextForDelivery(answer, artifacts), messageId);
       const delivery = await this.#deliverArtifacts(chatId, messageId, artifacts, textReceipt);
       const artifactDispatched = delivery.receipt.artifacts.some(
         ({ outcome }) => outcome === 'sent' || outcome === 'unknown',
@@ -5402,6 +5534,7 @@ export class FeishuHarnessBridge {
           completedAnswer = completed.answer;
           completedArtifacts = completed.artifacts ?? [];
           await controller.setContent(answerTextForDelivery(completedAnswer, completedArtifacts));
+          await this.#maybeSendVoiceReply(key, chatId, answerTextForDelivery(completedAnswer, completedArtifacts), messageId);
         },
       }, {
         replyTo: messageId,
@@ -5435,6 +5568,7 @@ export class FeishuHarnessBridge {
             fallbackError,
           );
         }
+        await this.#maybeSendVoiceReply(key, chatId, answerTextForDelivery(completedAnswer, completedArtifacts), messageId);
         const delivery = await this.#deliverArtifacts(
           chatId,
           messageId,
@@ -5491,6 +5625,7 @@ export class FeishuHarnessBridge {
           fallbackError,
         );
       }
+      await this.#maybeSendVoiceReply(key, chatId, answerTextForDelivery(answer, artifacts), messageId);
       const delivery = await this.#deliverArtifacts(chatId, messageId, artifacts, textReceipt);
       const artifactDispatched = delivery.receipt.artifacts.some(
         ({ outcome }) => outcome === 'sent' || outcome === 'unknown',
@@ -5670,6 +5805,8 @@ export class FeishuHarnessBridge {
     requiresMention,
     replyToMessageId,
   }) {
+    let approvalCardMessageId = null;
+    let approvalCardData = null;
     if (await this.#approvals.handleRequested(interaction, {
       key,
       actor,
@@ -5681,15 +5818,16 @@ export class FeishuHarnessBridge {
       ...(this.#interactionCards
         ? {
             render: async (pending) => {
-              // Show the approval as an interactive card with approve/reject buttons.
-              await this.#sendCard(
+              // Keep the original card details and id with this approval's callbacks.
+              approvalCardData = {
+                toolName: pending.toolCall?.name ?? pending.interaction.payload.toolName,
+                operation: operationArguments(pending.toolCall),
+                reason: pending.interaction.payload.reason,
+                approvalId: pending.approvalId,
+              };
+              approvalCardMessageId = await this.#sendCard(
                 chatId,
-                approvalCard({
-                  toolName: pending.toolCall?.name ?? pending.payload?.toolName,
-                  operation: operationArguments(pending.toolCall),
-                  reason: pending.payload?.reason,
-                  approvalId: pending.approvalId,
-                }),
+                approvalCard(approvalCardData),
                 { key, replyTo: replyToMessageId },
               ).catch(async () => {
                 // Fall back to the plain-text approval if the card cannot be
@@ -5697,8 +5835,17 @@ export class FeishuHarnessBridge {
                 // the pending approval is not marked as presented and the
                 // existing retry/reconnect logic can run.
                 await this.#send(chatId, pending.text, { replyTo: replyToMessageId });
+                return null;
               });
             },
+            onResolved: async (resolvedText) => {
+              if (!approvalCardMessageId) return;
+              await this.#patchCardMessage(
+                chatId,
+                approvalCardMessageId,
+                approvalCard({ ...approvalCardData, resolvedText }),
+              );
+            },
           }
         : {}),
     })) return;

package/src/channels/feishu/feishu-cards.mjs

@@ -877,7 +877,7 @@ export function customSteerCard() {
  * `requiresMention` is advisory; a button click is itself the operator's
  * explicit intent, so it does not need an @ mention in groups.
  */
-export function approvalCard({ toolName, operation, reason, approvalId }) {
+export function approvalCard({ toolName, operation, reason, approvalId, resolvedText }) {
   const elements = [];
   if (toolName) {
     elements.push({ tag: 'div', text: markdown(t('工具：{tool}', { tool: String(toolName) })) });
@@ -898,7 +898,9 @@ export function approvalCard({ toolName, operation, reason, approvalId }) {
   }
   elements.push(
     { tag: 'hr' },
-    buttonPair(t('✅ 批准'), `approve:${approvalId}`, t('❌ 拒绝'), `reject:${approvalId}`),
+    resolvedText
+      ? { tag: 'div', text: markdown(resolvedText) }
+      : buttonPair(t('✅ 批准'), `approve:${approvalId}`, t('❌ 拒绝'), `reject:${approvalId}`),
   );
   return cardWith(t('🔐 工具审批'), elements);
 }

package/src/channels/feishu/feishu-runtime.mjs

@@ -5,6 +5,7 @@ import { cardActionProbeCard } from './feishu-cards.mjs';
 import { VerifiedFeishuChannel } from './feishu-channel.mjs';
 import { normalizeFeishuGroupResponseMode } from './group-response-mode.mjs';
 import { normalizeFeishuStepPushMode } from './step-push-mode.mjs';
+import { createVoice } from './voice.mjs';
 import {
   registerSlashCommands,
   SLASH_COMMAND_MANIFEST,
@@ -115,6 +116,7 @@ export class FeishuRuntime {
   #groupTopicReply;
   #stepPush;
   #stepPushMode;
+  #voice = null;
   #sessionSyncTargetsFor;
   #ownerOpenIds;
   #harness;
@@ -149,6 +151,7 @@ export class FeishuRuntime {
     groupTopicReply = false,
     stepPush = false,
     stepPushMode = 'post',
+    voice = null,
     sessionSyncTargetsFor = null,
     ownerOpenId,
     ownerOpenIds,
@@ -188,6 +191,11 @@ export class FeishuRuntime {
     this.#groupTopicReply = groupTopicReply === true;
     this.#stepPush = stepPush === true;
     this.#stepPushMode = normalizeFeishuStepPushMode(stepPushMode);
+    // voice 为 { config, secret } 来源;凭据缺失时 createVoice 返回禁用对象,
+    // 桥接层完全跳过语音路径,不影响连接。
+    this.#voice = voice == null
+      ? null
+      : createVoice({ settings: voice.config, secret: voice.secret, logger });
     this.#sessionSyncTargetsFor = typeof sessionSyncTargetsFor === 'function'
       ? sessionSyncTargetsFor
       : null;
@@ -230,6 +238,13 @@ export class FeishuRuntime {
     this.#bridge?.setStepPushMode(this.#stepPushMode);
   }
 
+  setVoice(source) {
+    this.#voice = source == null
+      ? null
+      : createVoice({ settings: source.config, secret: source.secret, logger: this.#logger });
+    this.#bridge?.setVoice(this.#voice);
+  }
+
   async start() {
     while (true) {
       while (this.#stopping) await this.#stopping;
@@ -321,6 +336,7 @@ export class FeishuRuntime {
         groupTopicReply: this.#groupTopicReply,
         stepPush: this.#stepPush,
         stepPushMode: this.#stepPushMode,
+        voice: this.#voice,
         sessionSyncTargetsFor: this.#sessionSyncTargetsFor,
         repair: this.#repair,
         replyTimeoutMs: this.#replyTimeoutMs,

package/src/channels/feishu/multi-bot-controller.mjs

@@ -21,6 +21,7 @@ import {
   isFeishuStepPushMode,
   normalizeFeishuStepPushMode,
 } from './step-push-mode.mjs';
+import { normalizeFeishuVoiceConfig } from './voice-config.mjs';
 
 const ACTIVE_REGISTRATION_STATES = new Set([
   'starting', 'qr_ready', 'polling', 'slow_down', 'domain_switched',
@@ -99,6 +100,7 @@ function configuredBotFingerprint(config) {
     groupTopicReply: config.groupTopicReply === true,
     stepPush: config.stepPush === true,
     stepPushMode: normalizeFeishuStepPushMode(config.stepPushMode),
+    voice: config.voice,
     groupMessagePermissionGranted: config.groupMessagePermissionGranted === true,
     deletionPending: config.deletionPending === true,
     connectedAt: config.connectedAt ?? null,
@@ -645,6 +647,25 @@ export class MultiBotDshFeishuController {
     }));
   }
 
+  async updateVoice(botId, voice) {
+    this.#assertOpen();
+    if (voice !== null && (typeof voice !== 'object' || Array.isArray(voice))) {
+      throw new TypeError('Invalid Feishu voice configuration');
+    }
+    const normalized = normalizeFeishuVoiceConfig(voice);
+    return this.#serializeConfig(() => this.#withBotTransition(botId, async () => {
+      const config = this.#requireBot(botId);
+      const saved = await this.#configStore.saveBot({ ...config, voice: normalized });
+      // 语音是可选的渠道能力:凭据缺失时静默降级为禁用,不打断配置保存。
+      const secret = normalized
+        ? await atConnectionStage('credential.read', () => this.#credentials.resolve(normalized.secretRef), 'credential-store').catch(() => null)
+        : null;
+      this.#runtimes.get(botId)?.setVoice?.({ config: normalized, secret: secret?.value ?? null });
+      this.#touch();
+      return this.status(botId);
+    }));
+  }
+
   async deleteBot(botId) {
     this.#assertOpen();
     return this.#serializeConfig(() => this.#withBotTransition(botId, async () => {
@@ -710,6 +731,7 @@ export class MultiBotDshFeishuController {
         groupTopicReply: config.groupTopicReply === true,
         stepPush: config.stepPush === true,
         stepPushMode: normalizeFeishuStepPushMode(config.stepPushMode),
+        voice: normalizeFeishuVoiceConfig(config.voice),
         groupMessagePermissionGranted: config.groupMessagePermissionGranted === true,
         bot: publicBot(config),
         connection,

package/src/channels/feishu/plugin-config-store.mjs

@@ -3,6 +3,7 @@ import { mkdir, readFile, rename, unlink, writeFile } from 'node:fs/promises';
 import { dirname } from 'node:path';
 import { normalizeFeishuGroupResponseMode } from './group-response-mode.mjs';
 import { normalizeFeishuStepPushMode } from './step-push-mode.mjs';
+import { normalizeFeishuVoiceConfig } from './voice-config.mjs';
 
 export const LEGACY_FEISHU_SECRET_REF = 'DSH_FEISHU_APP_SECRET';
 
@@ -49,6 +50,7 @@ function normalizeBot(value, { legacy = false } = {}) {
     groupTopicReply: value.groupTopicReply === true,
     stepPush: value.stepPush === true,
     stepPushMode: normalizeFeishuStepPushMode(value.stepPushMode),
+    voice: normalizeFeishuVoiceConfig(value.voice),
     groupMessagePermissionGranted: value.groupMessagePermissionGranted === true,
     deletionPending: value.deletionPending === true,
     connectedAt: cleanString(value.connectedAt),

package/src/channels/feishu/voice-config.mjs

@@ -0,0 +1,34 @@
+// voice-config.mjs — 飞书语音配置的纯数据归一化(浏览器安全,无 Node 内置模块)。
+// 运行时转写/合成能力见 voice.mjs;此处只负责把每机器人配置校验为固定形状,
+// 供 host 侧配置存储、RPC 校验与 client 侧设置页共用。
+export const DEFAULT_VOICE_SECRET_REF = 'DASHSCOPE_API_KEY';
+export const DEFAULT_ASR_BASE_URL = 'https://dashscope.aliyuncs.com/compatible-mode/v1';
+export const DEFAULT_ASR_MODEL = 'qwen3-asr-flash';
+export const DEFAULT_TTS_MODEL = 'qwen3-tts-flash';
+export const DEFAULT_TTS_VOICE = 'Momo';
+export const TTS_ENDPOINT = 'https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation';
+export const MAX_TTS_CHARS = 600;
+
+function cleanString(value) {
+  return typeof value === 'string' && value.trim() ? value.trim() : null;
+}
+
+/**
+ * 归一化每机器人语音配置。语音是可选的渠道能力:关闭或配置不完整时返回 null,
+ * 与未配置行为完全一致。secretRef 走插件凭据库(与飞书 app secret 相同的环境变量
+ * 命名规范),密钥本身绝不进配置。
+ */
+export function normalizeFeishuVoiceConfig(value) {
+  if (value == null || typeof value !== 'object' || value.enabled !== true) return null;
+  const secretRef = cleanString(value.secretRef);
+  if (!secretRef || !/^[A-Za-z_][A-Za-z0-9_]*$/.test(secretRef)) return null;
+  return Object.freeze({
+    enabled: true,
+    secretRef,
+    asrModel: cleanString(value.asrModel) ?? DEFAULT_ASR_MODEL,
+    ttsModel: cleanString(value.ttsModel) ?? DEFAULT_TTS_MODEL,
+    ttsVoice: cleanString(value.ttsVoice) ?? DEFAULT_TTS_VOICE,
+    asrBaseUrl: cleanString(value.asrBaseUrl) ?? DEFAULT_ASR_BASE_URL,
+    ffmpeg: cleanString(value.ffmpeg),
+  });
+}

package/src/channels/feishu/voice.mjs

@@ -0,0 +1,134 @@
+// voice.mjs — 飞书渠道语音能力运行时(渠道能力:语音输入与语音回复)
+// 入站:音频消息 opus → ffmpeg 转 wav 16k → DashScope qwen3-asr-flash 转文字,
+//       回流入正常文本流水线;转写失败走明确降级(保持原有"仅支持文字/图片/文件"提示)。
+// 出站:同回合文字回复 → qwen3-tts-flash → ffmpeg 转 opus → 飞书音频消息,
+//       优先回复原语音消息,reply 不可用时降级为普通音频消息。
+// 凭据经插件凭据库(secretRef)在运行时注入;本模块不落盘、不把密钥写进日志。
+// 配置归一化(浏览器安全)在 voice-config.mjs,供 client 侧设置页与 host 侧共用。
+import { spawn } from 'node:child_process';
+import { randomUUID } from 'node:crypto';
+import { mkdirSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { MAX_TTS_CHARS, TTS_ENDPOINT, normalizeFeishuVoiceConfig } from './voice-config.mjs';
+
+function runFfmpeg(bin, args) {
+  return new Promise((resolve, reject) => {
+    const p = spawn(bin, ['-hide_banner', '-loglevel', 'error', '-y', ...args]);
+    let err = '';
+    p.stderr.on('data', (d) => { err += d; if (err.length > 2000) err = err.slice(-2000); });
+    // 启动失败(未安装 ffmpeg、配置路径不存在或不可执行)时 spawn 触发
+    // 'error' 事件而非 'close';不监听会让未处理的 ENOENT 异常击穿 Node 进程。
+    // 这里改为正常 reject,由外层的转写/合成失败降级统一兜底。
+    p.on('error', reject);
+    p.on('close', (code) => (code === 0 ? resolve() : reject(new Error(`ffmpeg exit ${code}: ${err.slice(-300)}`))));
+  });
+}
+
+async function readResourceToBuffer(resource) {
+  if (typeof resource?.writeFile === 'function') {
+    const path = join(tmpdir(), `dsh-voice-in-${randomUUID()}.opus`);
+    await resource.writeFile(path);
+    const buf = readFileSync(path);
+    rmSync(path, { force: true });
+    return buf;
+  }
+  if (resource?.getReadableStream) {
+    const chunks = [];
+    for await (const chunk of resource.getReadableStream()) {
+      chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk));
+    }
+    return Buffer.concat(chunks);
+  }
+  throw new Error('消息资源无可用读取接口');
+}
+
+/**
+ * 创建语音能力对象。settings 为 normalizeFeishuVoiceConfig 的产物,secret 为
+ * 凭据库解析出的 DashScope api_key。两者任一缺失即返回 { enabled: false },
+ * 桥接层据此完全跳过语音路径。
+ */
+export function createVoice({ settings, secret, logger } = {}) {
+  const config = normalizeFeishuVoiceConfig(settings);
+  const apiKey = typeof secret === 'string' && secret.trim() ? secret.trim() : null;
+  if (!config || !apiKey) {
+    if (config) logger?.warn?.('[dsh-feishu-voice] 语音凭据缺失,语音已禁用');
+    return Object.freeze({ enabled: false });
+  }
+  const frozen = Object.freeze({
+    apiKey,
+    asrBaseUrl: config.asrBaseUrl,
+    asrModel: config.asrModel,
+    ttsModel: config.ttsModel,
+    ttsVoice: config.ttsVoice,
+    ffmpeg: config.ffmpeg ?? 'ffmpeg',
+  });
+
+  async function transcribeIncoming(event, client) {
+    const fileKey = JSON.parse(event.message.content).file_key;
+    if (!fileKey) return null;
+    const resource = await client.im.v1.messageResource.get({
+      params: { type: 'file' },
+      path: { message_id: event.message.message_id, file_key: fileKey },
+    });
+    const opusBuf = await readResourceToBuffer(resource);
+    const work = join(tmpdir(), `dsh-voice-${randomUUID()}`);
+    mkdirSync(work, { recursive: true });
+    try {
+      const opusIn = join(work, 'in.opus');
+      const wavIn = join(work, 'in.wav');
+      writeFileSync(opusIn, opusBuf);
+      await runFfmpeg(frozen.ffmpeg, ['-i', opusIn, '-ar', '16000', '-ac', '1', wavIn]);
+      const b64 = readFileSync(wavIn).toString('base64');
+      // qwen3-asr-flash 走 compatible-mode chat/completions + input_audio data URI
+      const res = await fetch(`${frozen.asrBaseUrl}/chat/completions`, {
+        method: 'POST',
+        headers: { Authorization: `Bearer ${frozen.apiKey}`, 'Content-Type': 'application/json' },
+        body: JSON.stringify({
+          model: frozen.asrModel,
+          messages: [{
+            role: 'user',
+            content: [{ type: 'input_audio', input_audio: { data: `data:audio/wav;base64,${b64}` } }],
+          }],
+        }),
+      });
+      const text = await res.text();
+      if (!res.ok) throw new Error(`dashscope ASR ${res.status}: ${text.slice(0, 200)}`);
+      const j = JSON.parse(text);
+      const out = j?.choices?.[0]?.message?.content;
+      if (typeof out !== 'string' || !out.trim()) throw new Error('ASR 无文本输出');
+      return out.trim();
+    } finally {
+      rmSync(work, { recursive: true, force: true });
+    }
+  }
+
+  async function synthesize(text) {
+    const clean = String(text ?? '').replace(/\s+/g, ' ').trim().slice(0, MAX_TTS_CHARS);
+    if (!clean) return null;
+    // qwen3-tts-flash 走 DashScope 原生 multimodal-generation 接口(compatible-mode 不支持 TTS)
+    const res = await fetch(TTS_ENDPOINT, {
+      method: 'POST',
+      headers: { Authorization: `Bearer ${frozen.apiKey}`, 'Content-Type': 'application/json' },
+      body: JSON.stringify({ model: frozen.ttsModel, input: { text: clean, voice: frozen.ttsVoice } }),
+    });
+    const j = await res.json().catch(() => null);
+    const url = j?.output?.audio?.url;
+    if (!url) throw new Error(`TTS 无音频:${res.status} ${JSON.stringify(j ?? {}).slice(0, 160)}`);
+    const aud = await fetch(url);
+    if (!aud.ok) throw new Error(`TTS 音频下载 ${aud.status}`);
+    const work = join(tmpdir(), `dsh-voice-${randomUUID()}`);
+    mkdirSync(work, { recursive: true });
+    try {
+      const wavOut = join(work, 'out.wav');
+      const opusOut = join(work, 'out.opus');
+      writeFileSync(wavOut, Buffer.from(await aud.arrayBuffer()));
+      await runFfmpeg(frozen.ffmpeg, ['-i', wavOut, '-c:a', 'libopus', '-b:a', '48k', opusOut]);
+      return readFileSync(opusOut);
+    } finally {
+      rmSync(work, { recursive: true, force: true });
+    }
+  }
+
+  return Object.freeze({ enabled: true, transcribeIncoming, synthesize });
+}