botium-connector-voip 0.0.19 → 0.0.21

package/dist/botium-connector-voip-es.js

@@ -1,8 +1,9 @@
 import ws from 'ws';
 import lodash from 'lodash';
 import axios from 'axios';
+import http from 'http';
+import https from 'https';
 import debug$4 from 'debug';
-import path from 'path';
 import musicMetadata from 'music-metadata';
 
 const debug$3 = debug$4('botium-connector-voip');
@@ -24,6 +25,9 @@ const Capabilities = {
   VOIP_TTS_BODY: 'VOIP_TTS_BODY',
   VOIP_TTS_HEADERS: 'VOIP_TTS_HEADERS',
   VOIP_TTS_TIMEOUT: 'VOIP_TTS_TIMEOUT',
+  VOIP_TTS_CACHE_ENABLE: 'VOIP_TTS_CACHE_ENABLE',
+  VOIP_TTS_CACHE_SIZE: 'VOIP_TTS_CACHE_SIZE',
+  VOIP_TTS_PREFETCH_ENABLE: 'VOIP_TTS_PREFETCH_ENABLE',
   VOIP_WORKER_URL: 'VOIP_WORKER_URL',
   VOIP_WORKER_APIKEY: 'VOIP_WORKER_APIKEY',
   VOIP_SIP_POOL_CALLER_ENABLE: 'VOIP_SIP_POOL_CALLER_ENABLE',
@@ -57,6 +61,9 @@ const Defaults = {
   VOIP_STT_TIMEOUT: 10000,
   VOIP_TTS_METHOD: 'GET',
   VOIP_TTS_TIMEOUT: 10000,
+  VOIP_TTS_CACHE_ENABLE: true,
+  VOIP_TTS_CACHE_SIZE: 50,
+  VOIP_TTS_PREFETCH_ENABLE: true,
   VOIP_STT_MESSAGE_HANDLING: 'ORIGINAL',
   VOIP_STT_MESSAGE_HANDLING_TIMEOUT: 2500,
   VOIP_STT_MESSAGE_HANDLING_DELIMITER: '. ',
@@ -71,6 +78,12 @@ const Defaults = {
   VOIP_USE_GLOBAL_VOIP_WORKER: false,
   VOIP_SIP_PROTOCOL: 'TCP'
 };
+const TTS_HTTP_AGENT = new http.Agent({
+  keepAlive: true
+});
+const TTS_HTTPS_AGENT = new https.Agent({
+  keepAlive: true
+});
 class BotiumConnectorVoip {
   constructor({
     queueBotSays,
@@ -84,9 +97,18 @@ class BotiumConnectorVoip {
     this.sentencesBuilding = 0;
     this.sentencesFinal = 0;
     this.sentenceBuilding = false;
+    this.ttsCache = new Map();
+    this.ttsCacheEnabled = false;
+    this.ttsCacheMaxEntries = 0;
+    this.ttsPrefetchEnabled = false;
+
+    // For debugging latency between incoming STT (bot says) and outgoing audio (sendAudio)
+    this._lastBotSaysQueuedAt = null;
+    this._lastBotSaysText = null;
   }
   async Validate() {
     debug$3('Validate called');
+    this.caps = Object.assign({}, Defaults, this.caps);
     debug$3(this.caps.VOIP_STT_MESSAGE_HANDLING);
     if (this.caps.VOIP_TTS_URL) {
       this.axiosTtsParams = {
@@ -94,7 +116,9 @@ class BotiumConnectorVoip {
         params: this._getParams(Capabilities.VOIP_TTS_PARAMS),
         method: this.caps.VOIP_TTS_METHOD,
         timeout: this.caps.VOIP_TTS_TIMEOUT,
-        headers: this._getHeaders(Capabilities.VOIP_TTS_HEADERS)
+        headers: this._getHeaders(Capabilities.VOIP_TTS_HEADERS),
+        httpAgent: TTS_HTTP_AGENT,
+        httpsAgent: TTS_HTTPS_AGENT
       };
       try {
         const {
@@ -112,7 +136,10 @@ class BotiumConnectorVoip {
         throw new Error(`Checking TTS Status failed - ${this._getAxiosErrOutput(err)}`);
       }
     }
-    this.caps = Object.assign({}, Defaults, this.caps);
+    const cacheSize = parseInt(this.caps[Capabilities.VOIP_TTS_CACHE_SIZE], 10);
+    this.ttsCacheEnabled = !!this.caps[Capabilities.VOIP_TTS_CACHE_ENABLE];
+    this.ttsCacheMaxEntries = lodash.isFinite(cacheSize) && cacheSize > 0 ? cacheSize : 0;
+    this.ttsPrefetchEnabled = !!this.caps[Capabilities.VOIP_TTS_PREFETCH_ENABLE];
   }
   async Start() {
     debug$3('Start called');
@@ -122,19 +149,31 @@ class BotiumConnectorVoip {
     this.end = false;
     this.connected = false;
     this.convoStep = null;
+    this._lastBotSaysQueuedAt = null;
+    this._lastBotSaysText = null;
+    if (this.ttsCache) {
+      this.ttsCache.clear();
+    }
     const sendBotMsg = botMsg => {
+      this._lastBotSaysQueuedAt = Date.now();
+      this._lastBotSaysText = botMsg && botMsg.messageText ? String(botMsg.messageText) : null;
       setTimeout(() => this.queueBotSays(botMsg), 0);
     };
     const joinBotMsg = (botMsgs, joinLastPrevMsg) => {
       const botMsg = {};
       botMsg.messageText = botMsgs.map(m => m.messageText).join(this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_DELIMITER] || '');
       botMsg.sourceData = botMsgs.map(m => m.sourceData);
-      if (lodash.isNil(joinLastPrevMsg)) {
-        botMsg.sourceData[0].silenceDuration = botMsgs[0].sourceData.data.start;
-        botMsg.sourceData[0].voiceDuration = botMsgs[botMsgs.length - 1].sourceData.data.end - botMsgs[0].sourceData.data.start;
+      const firstStart = lodash.get(botMsgs, '[0].sourceData.data.start', null);
+      const lastEnd = lodash.get(botMsgs, `[${botMsgs.length - 1}].sourceData.data.end`, null);
+      const prevEnd = lodash.get(joinLastPrevMsg, 'sourceData.data.end', null);
+      // joinLastPrevMsg can be null (first joined message) OR incomplete (in case JOIN timeout fired before we got any final STT info).
+      // In those cases, fall back to "start-of-first-chunk" semantics (same as previous behaviour for nil joinLastPrevMsg).
+      if (lodash.isFinite(firstStart) && lodash.isFinite(lastEnd)) {
+        botMsg.sourceData[0].silenceDuration = lodash.isFinite(prevEnd) ? firstStart - prevEnd : firstStart;
+        botMsg.sourceData[0].voiceDuration = lastEnd - firstStart;
       } else {
-        botMsg.sourceData[0].silenceDuration = botMsgs[0].sourceData.data.start - joinLastPrevMsg.sourceData.data.end;
-        botMsg.sourceData[0].voiceDuration = botMsgs[botMsgs.length - 1].sourceData.data.end - botMsgs[0].sourceData.data.start;
+        botMsg.sourceData[0].silenceDuration = lodash.isFinite(firstStart) ? firstStart : null;
+        botMsg.sourceData[0].voiceDuration = lodash.isFinite(firstStart) && lodash.isFinite(lastEnd) ? lastEnd - firstStart : null;
       }
       return botMsg;
     };
@@ -163,12 +202,14 @@ class BotiumConnectorVoip {
     const connect = async retryIndex => {
       retryIndex = retryIndex || 0;
       try {
+        const workerUrl = this.caps[Capabilities.VOIP_USE_GLOBAL_VOIP_WORKER] ? process.env.BOTIUM_VOIP_WORKER_URL : this.caps[Capabilities.VOIP_WORKER_URL].replace('wss', 'https').replace('ws', 'http');
+        const baseUrl = workerUrl.endsWith('/') ? workerUrl.slice(0, -1) : workerUrl;
         const res = await axios({
           method: 'post',
           data: {
             API_KEY: this.caps[Capabilities.VOIP_USE_GLOBAL_VOIP_WORKER] ? process.env.BOTIUM_VOIP_WORKER_APIKEY : this.caps[Capabilities.VOIP_WORKER_APIKEY]
           },
-          url: new URL(path.join(`${this.caps[Capabilities.VOIP_USE_GLOBAL_VOIP_WORKER] ? process.env.BOTIUM_VOIP_WORKER_URL : this.caps[Capabilities.VOIP_WORKER_URL].replace('wss', 'https').replace('ws', 'http')}`, 'initCall')).toString()
+          url: `${baseUrl}/initCall`
         });
         if (res) {
           data = res.data;
@@ -222,6 +263,35 @@ class BotiumConnectorVoip {
         }
         this.eventEmitter.on('CONVO_STEP_NEXT', (container, convoStep) => {
           this.convoStep = convoStep;
+          this._maybePrefetchTts(convoStep);
+          // For PSST: send join silence duration per step to VOIP worker (controls PSST silence trigger)
+          try {
+            if (this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'PSST' && this.ws && this.ws.readyState === ws.OPEN) {
+              const joinLogicHook = this._getJoinLogicHook(convoStep);
+              let silenceMs = null;
+              if (joinLogicHook && joinLogicHook.args && joinLogicHook.args.length > 0) {
+                silenceMs = parseInt(joinLogicHook.args[0], 10);
+              }
+              // Fallback to global timeout if no per-step hook is set
+              if (!lodash.isFinite(silenceMs) || silenceMs <= 0) {
+                silenceMs = parseInt(this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT], 10);
+              }
+              // PSST: treat like JOIN, but subtract 500ms from timeout
+              if (lodash.isFinite(silenceMs) && silenceMs > 0) {
+                silenceMs = Math.max(0, silenceMs - 500);
+              }
+              if (lodash.isFinite(silenceMs) && silenceMs > 0 && this.sessionId) {
+                debug$3(`PSST: sending silenceDurationMs=${silenceMs} for sessionId=${this.sessionId}`);
+                this.ws.send(JSON.stringify({
+                  METHOD: 'setSttSilenceDuration',
+                  sessionId: this.sessionId,
+                  silenceDurationMs: silenceMs
+                }));
+              }
+            }
+          } catch (err) {
+            debug$3(`Failed sending PSST silence duration to VOIP worker: ${err.message || err}`);
+          }
         });
         this.silence = null;
         this.msgCount = 0;
@@ -230,6 +300,18 @@ class BotiumConnectorVoip {
         this.silenceTimeout = null;
         this.wsOpened = true;
         debug$3(`Websocket connection to ${wsEndpoint} opened.`);
+        const sttHandling = this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING];
+        const isPsst = sttHandling === 'PSST';
+        const sttLegacy = true;
+        let sttUrl = this.caps[Capabilities.VOIP_STT_URL_STREAM];
+        if (isPsst && lodash.isString(sttUrl)) {
+          const replaced = sttUrl.replace('/api/sttstream/', '/api/stt/');
+          if (replaced !== sttUrl) {
+            sttUrl = replaced;
+          } else {
+            debug$3(`PSST: Could not derive /api/stt url from ${sttUrl}, using as-is`);
+          }
+        }
         const request = {
           METHOD: 'initCall',
           SIP_CALLER_AUTO: this.caps[Capabilities.VOIP_SIP_POOL_CALLER_ENABLE],
@@ -250,8 +332,9 @@ class BotiumConnectorVoip {
           ICE_TURN_PASSWORD: this.caps[Capabilities.VOIP_ICE_TURN_PASSWORD],
           ICE_TURN_PROTOCOL: this.caps[Capabilities.VOIP_ICE_TURN_PROTOCOL] || 'TCP',
           MIN_SILENCE_DURATION: this.caps[Capabilities.VOIP_SILENCE_DURATION_TIMEOUT_ENABLE] ? this.caps[Capabilities.VOIP_SILENCE_DURATION_TIMEOUT] : null,
+          STT_LEGACY: sttLegacy,
           STT_CONFIG: {
-            stt_url: this.caps[Capabilities.VOIP_STT_URL_STREAM],
+            stt_url: sttUrl,
             stt_params: this.caps[Capabilities.VOIP_STT_PARAMS_STREAM],
             stt_body: this.caps[Capabilities.VOIP_STT_BODY_STREAM] || null
           },
@@ -272,9 +355,88 @@ class BotiumConnectorVoip {
         });
         this.ws.on('message', async data => {
           const parsedData = JSON.parse(data);
+          // Allow fullRecord delivery even if Stop() was already called.
+          // Otherwise the recording can be dropped if the worker streams it late (common on hangup).
+          if (this.stopCalled) {
+            const allowedTypes = ['fullRecord', 'fullRecordStart', 'fullRecordChunk', 'fullRecordEnd', 'error'];
+            if (!parsedData || !parsedData.type || !allowedTypes.includes(parsedData.type)) {
+              debug$3(`${this.sessionId} - Stop already called, ignoring incoming message`);
+              return;
+            }
+          }
           const parsedDataLog = lodash.cloneDeep(parsedData);
-          parsedDataLog.fullRecord = '<full_record_buffer>';
+          // Sanitize all potential base64 fields for logging (don't dump huge buffers)
+          const sanitizeBase64Fields = (obj, prefix = '') => {
+            if (!obj || typeof obj !== 'object') return;
+            for (const key of Object.keys(obj)) {
+              const val = obj[key];
+              if (typeof val === 'string' && val.length > 500) {
+                obj[key] = `<base64:${val.length}chars>`;
+              } else if (val && typeof val === 'object' && !Array.isArray(val)) {
+                sanitizeBase64Fields(val, `${prefix}${key}.`);
+              }
+            }
+          };
+          sanitizeBase64Fields(parsedDataLog);
           debug$3(JSON.stringify(parsedDataLog, null, 2));
+          const _extractFullRecordBase64 = pd => {
+            if (!pd) return null;
+            // Different VOIP workers may put the payload in various fields - search all string fields
+            // First try known field names
+            const knownFields = [pd.fullRecord, pd.full_record, pd.data?.fullRecord, pd.data?.full_record, pd.data?.b64_buffer, pd.data?.base64, pd.data?.full_record_buffer, pd.data?.buffer, pd.buffer, pd.base64, pd.audio, pd.audioData, pd.data?.audio, pd.data?.audioData];
+            for (const val of knownFields) {
+              if (typeof val === 'string' && val.length > 100) return val;
+            }
+            // Fallback: find any large string field (likely base64)
+            const findLargeString = (obj, depth = 0) => {
+              if (!obj || typeof obj !== 'object' || depth > 3) return null;
+              for (const key of Object.keys(obj)) {
+                const val = obj[key];
+                if (typeof val === 'string' && val.length > 1000) {
+                  debug$3(`${this.sessionId} - Found base64 in field '${key}' (len=${val.length})`);
+                  return val;
+                }
+                if (val && typeof val === 'object' && !Array.isArray(val)) {
+                  const found = findLargeString(val, depth + 1);
+                  if (found) return found;
+                }
+              }
+              return null;
+            };
+            return findLargeString(pd);
+          };
+          const emitFullRecordAttachment = base64 => {
+            if (!base64 || typeof base64 !== 'string' || base64.length === 0) {
+              // Log available fields (length only) to help diagnose worker payload shape without dumping base64
+              try {
+                const candidates = {
+                  fullRecord: typeof parsedData?.fullRecord === 'string' ? parsedData.fullRecord.length : null,
+                  full_record: typeof parsedData?.full_record === 'string' ? parsedData.full_record.length : null,
+                  data_fullRecord: typeof parsedData?.data?.fullRecord === 'string' ? parsedData.data.fullRecord.length : null,
+                  data_full_record: typeof parsedData?.data?.full_record === 'string' ? parsedData.data.full_record.length : null,
+                  data_b64_buffer: typeof parsedData?.data?.b64_buffer === 'string' ? parsedData.data.b64_buffer.length : null,
+                  data_base64: typeof parsedData?.data?.base64 === 'string' ? parsedData.data.base64.length : null,
+                  data_full_record_buffer: typeof parsedData?.data?.full_record_buffer === 'string' ? parsedData.data.full_record_buffer.length : null,
+                  data_buffer: typeof parsedData?.data?.buffer === 'string' ? parsedData.data.buffer.length : null
+                };
+                debug$3(`${this.sessionId} - fullRecordEnd received but base64 is empty, skipping attachment emit. CandidateLengths=${JSON.stringify(candidates)}`);
+              } catch (err) {
+                debug$3(`${this.sessionId} - fullRecordEnd received but base64 is empty, skipping attachment emit (diag failed: ${err.message || err})`);
+              }
+              return;
+            }
+            debug$3(`${this.sessionId} - Emitting MESSAGE_ATTACHMENT full_record.wav (base64Len=${base64.length}, stopCalled=${!!this.stopCalled})`);
+            this.eventEmitter.emit('MESSAGE_ATTACHMENT', this.container, {
+              name: 'full_record.wav',
+              mimeType: 'audio/wav',
+              base64,
+              // Session context from capabilities for correlation
+              sessionContext: {
+                testSessionId: this.caps.VOIP_TEST_SESSION_ID || null,
+                testSessionJobId: this.caps.VOIP_TEST_SESSION_JOB_ID || null
+              }
+            });
+          };
           if (parsedData && parsedData.type === 'callinfo' && parsedData.status === 'initialized') {
             this.sessionId = parsedData.voipConfig.sessionId;
           }
@@ -310,6 +472,22 @@ class BotiumConnectorVoip {
             reject(new Error(`Error: ${parsedData.message}`));
             sendBotMsg(new Error(`Error: ${parsedData.message}`));
           }
+
+          // Full record streaming support:
+          // - some VOIP workers send the recording in chunks and an end marker
+          if (parsedData && parsedData.type === 'fullRecordStart') {
+            this.fullRecord = '';
+          }
+          if (parsedData && parsedData.type === 'fullRecordChunk') {
+            const chunk = _extractFullRecordBase64(parsedData);
+            if (chunk) this.fullRecord = (this.fullRecord || '') + chunk;
+          }
+          if (parsedData && parsedData.type === 'fullRecordEnd') {
+            const tail = _extractFullRecordBase64(parsedData);
+            if (tail) this.fullRecord = (this.fullRecord || '') + tail;
+            emitFullRecordAttachment(this.fullRecord || tail);
+            this.end = true;
+          }
           if (parsedData && parsedData.type === 'silence') {
             if (lodash.isNil(this._getIgnoreSilenceDurationAsserterLogicHook(this.convoStep))) {
               if (lodash.isNil(this._getJoinLogicHook(this.convoStep)) && parsedData.data.silence.length > 0) {
@@ -319,11 +497,8 @@ class BotiumConnectorVoip {
             }
           }
           if (parsedData && parsedData.type === 'fullRecord') {
-            this.eventEmitter.emit('MESSAGE_ATTACHMENT', this.container, {
-              name: 'full_record.wav',
-              mimeType: 'audio/wav',
-              base64: parsedData.fullRecord
-            });
+            // Non-chunked full record
+            emitFullRecordAttachment(_extractFullRecordBase64(parsedData));
             this.end = true;
           }
           if (parsedData && parsedData.data && parsedData.data.final === false) {
@@ -337,16 +512,29 @@ class BotiumConnectorVoip {
               }
             }
             this.firstSttInfoReceived = true;
-            if (this.prevData) {
+            if (this.prevData && this.prevData.data && !lodash.isNil(this.prevData.data.end)) {
               if (!lodash.isNil(parsedData.data.start)) {
                 const silenceDuration = parsedData.data.start - this.prevData.data.end;
                 const joinLogicHook = this._getJoinLogicHook(this.convoStep);
-                const isJoinMethod = this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'JOIN';
+                const sttHandling = this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING];
+                const isPsst = sttHandling === 'PSST';
+                const isJoinMethod = sttHandling === 'JOIN' || isPsst;
+                const toJoinTimeoutMs = ms => {
+                  const parsed = parseInt(ms, 10);
+                  if (!lodash.isFinite(parsed) || parsed <= 0) return null;
+                  return isPsst ? Math.max(0, parsed - 500) : parsed;
+                };
                 let matched = false;
-                if ((!lodash.isNil(joinLogicHook) && isJoinMethod && silenceDuration > parseInt(joinLogicHook.args[0])) / 1000) {
-                  matched = true;
-                } else if (lodash.isNil(joinLogicHook) && isJoinMethod && silenceDuration > parseInt(this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT]) / 1000) {
-                  matched = true;
+                if (!lodash.isNil(joinLogicHook) && isJoinMethod) {
+                  const joinTimeoutMs = toJoinTimeoutMs(joinLogicHook && joinLogicHook.args && joinLogicHook.args[0]);
+                  if (lodash.isFinite(joinTimeoutMs) && joinTimeoutMs > 0 && silenceDuration > joinTimeoutMs / 1000) {
+                    matched = true;
+                  }
+                } else if (lodash.isNil(joinLogicHook) && isJoinMethod) {
+                  const joinTimeoutMs = toJoinTimeoutMs(this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT]);
+                  if (lodash.isFinite(joinTimeoutMs) && joinTimeoutMs > 0 && silenceDuration > joinTimeoutMs / 1000) {
+                    matched = true;
+                  }
                 }
                 if (matched && this.botMsgs.length > 0) {
                   sendBotMsg(joinBotMsg(this.botMsgs, this.joinLastPrevMsg));
@@ -361,6 +549,7 @@ class BotiumConnectorVoip {
             const confidenceThreshold = this._getConfidenceScoreLogicHook(this.convoStep) && this._getConfidenceScoreLogicHook(this.convoStep).args[0] || this.caps[Capabilities.VOIP_STT_CONFIDENCE_THRESHOLD];
             const successfulConfidenceScore = this._getConfidenceScore(parsedData) >= confidenceThreshold;
             debug$3(`Message: ${parsedData.data.message} / Confidence Score: ${this._getConfidenceScore(parsedData)} (Threshold: ${confidenceThreshold})`);
+            // ORIGINAL: emit final message immediately, unless join hooks are active.
             if (this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'ORIGINAL' && lodash.isNil(this._getJoinLogicHook(this.convoStep))) {
               let botMsg = {
                 messageText: parsedData.data.message
@@ -375,7 +564,9 @@ class BotiumConnectorVoip {
                 this.firstMsg = false;
               } else {
                 const sourceData = parsedData;
-                sourceData.silenceDuration = parsedData.data.start - this.prevData.data.end;
+                const start = lodash.get(parsedData, 'data.start', null);
+                const prevEnd = lodash.get(this.prevData, 'data.end', null);
+                sourceData.silenceDuration = lodash.isFinite(start) && lodash.isFinite(prevEnd) ? start - prevEnd : lodash.isFinite(start) ? start : null;
                 sourceData.voiceDuration = parsedData.data.end - parsedData.data.start;
                 botMsg = Object.assign({}, botMsg, {
                   sourceData
@@ -402,7 +593,9 @@ class BotiumConnectorVoip {
                 this.firstMsg = false;
               } else {
                 const sourceData = parsedData;
-                sourceData.silenceDuration = parsedData.data.start - this.prevData.data.end;
+                const start = lodash.get(parsedData, 'data.start', null);
+                const prevEnd = lodash.get(this.prevData, 'data.end', null);
+                sourceData.silenceDuration = lodash.isFinite(start) && lodash.isFinite(prevEnd) ? start - prevEnd : lodash.isFinite(start) ? start : null;
                 sourceData.voiceDuration = parsedData.data.end - parsedData.data.start;
                 botMsg = Object.assign({}, botMsg, {
                   sourceData
@@ -416,7 +609,7 @@ class BotiumConnectorVoip {
               }
               this.botMsgs = [];
             }
-            if (this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'JOIN' || this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'CONCAT' || !lodash.isNil(this._getJoinLogicHook(this.convoStep))) {
+            if (this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'JOIN' || this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'PSST' || this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING] === 'CONCAT' || !lodash.isNil(this._getJoinLogicHook(this.convoStep))) {
               const botMsg = {
                 messageText: parsedData.data.message,
                 sourceData: parsedData
@@ -424,15 +617,27 @@ class BotiumConnectorVoip {
               this.prevData = parsedData;
               if (successfulConfidenceScore) {
                 this.botMsgs.push(botMsg);
+                const sttHandling = this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING];
+                const isPsst = sttHandling === 'PSST';
+                const toJoinTimeoutMs = ms => {
+                  const parsed = parseInt(ms, 10);
+                  if (!lodash.isFinite(parsed) || parsed <= 0) return null;
+                  return isPsst ? Math.max(0, parsed - 500) : parsed;
+                };
+                const joinLogicHook = this._getJoinLogicHook(this.convoStep);
+                let joinTimeoutMs = toJoinTimeoutMs(lodash.get(joinLogicHook, 'args[0]'));
+                if (!lodash.isFinite(joinTimeoutMs) || joinTimeoutMs <= 0) {
+                  joinTimeoutMs = toJoinTimeoutMs(this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT]);
+                }
                 this.silenceTimeout = setTimeout(() => {
                   if (this.botMsgs.length > 0) {
-                    debug$3('Silence Duration Timeout (PSST):', this._getJoinLogicHook(this.convoStep) && parseInt(this._getJoinLogicHook(this.convoStep).args[0]) || this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT], 'ms');
+                    debug$3('Silence Duration Timeout (JOIN/PSST):', joinTimeoutMs, 'ms');
                     sendBotMsg(joinBotMsg(this.botMsgs, this.joinLastPrevMsg));
                     this.firstMsg = false;
                     this.joinLastPrevMsg = this.botMsgs[this.botMsgs.length - 1];
                     this.botMsgs = [];
                   }
-                }, this._getJoinLogicHook(this.convoStep) && parseInt(this._getJoinLogicHook(this.convoStep).args[0]) || this.caps[Capabilities.VOIP_STT_MESSAGE_HANDLING_TIMEOUT]);
+                }, joinTimeoutMs || 0);
               }
             }
           }
@@ -444,7 +649,24 @@ class BotiumConnectorVoip {
   }
   async UserSays(msg) {
     debug$3('UserSays called');
-    debug$3(msg);
+    // Avoid logging large buffers/base64 (can break job logs and overwhelm stdout)
+    try {
+      const safeLog = {
+        messageText: msg && msg.messageText ? String(msg.messageText).substring(0, 200) : undefined,
+        buttons: msg && Array.isArray(msg.buttons) ? msg.buttons.length : 0,
+        hasMedia: !!(msg && msg.media && msg.media.length > 0),
+        mediaMimeType: msg && msg.media && msg.media[0] ? msg.media[0].mimeType : undefined,
+        mediaSize: msg && msg.media && msg.media[0] && Buffer.isBuffer(msg.media[0].buffer) ? msg.media[0].buffer.length : undefined,
+        attachments: msg && Array.isArray(msg.attachments) ? msg.attachments.map(a => ({
+          name: a && (a.name || a.title),
+          mimeType: a && a.mimeType,
+          base64Len: a && a.base64 ? a.base64.length : 0
+        })) : []
+      };
+      debug$3(safeLog);
+    } catch (err) {
+      debug$3(`UserSays: failed to build safe log: ${err.message || err}`);
+    }
     if (!msg.attachments) {
       msg.attachments = [];
     }
@@ -459,23 +681,32 @@ class BotiumConnectorVoip {
           });
           this.ws.send(request);
         } else if (msg && msg.messageText) {
-          if (!this.axiosTtsParams) reject(new Error('TTS not configured, only audio input supported'));
-          if (this.axiosTtsParams) {
-            const ttsRequest = {
-              ...this.axiosTtsParams,
-              params: {
-                ...(this.axiosTtsParams.params || {}),
-                text: msg.messageText
-              },
-              data: this._getBody(Capabilities.VOIP_TTS_BODY),
-              responseType: 'arraybuffer'
-            };
+          // Check for DTMF tag in messageText: <DTMF>1234</DTMF>
+          const dtmfMatch = msg.messageText.match(/<DTMF>([^<]+)<\/DTMF>/i);
+          if (dtmfMatch && dtmfMatch[1]) {
+            const digits = dtmfMatch[1];
+            debug$3(`Sending DTMF from messageText: ${digits}`);
+            const request = JSON.stringify({
+              METHOD: 'sendDtmf',
+              digits,
+              sessionId: this.sessionId
+            });
+            this.ws.send(request);
+            return resolve();
+          }
+          if (!this.axiosTtsParams) {
+            if (!(msg.media && msg.media.length > 0 && msg.media[0].buffer)) {
+              return reject(new Error('TTS not configured, only audio input supported'));
+            }
+          } else {
+            const ttsRequest = this._buildTtsRequest(msg.messageText);
+            if (!ttsRequest) return reject(new Error('TTS not configured, only audio input supported'));
             msg.sourceData = ttsRequest;
-            let ttsResponse = null;
+            let ttsResult = null;
             try {
-              ttsResponse = await axios(ttsRequest);
+              ttsResult = await this._getTtsAudio(ttsRequest, msg.messageText);
             } catch (err) {
-              reject(new Error(`TTS "${msg.messageText}" failed - ${this._getAxiosErrOutput(err)}`));
+              return reject(new Error(`TTS "${msg.messageText}" failed - ${this._getAxiosErrOutput(err)}`));
             }
             if (msg && msg.messageText && msg.messageText.length > 0) {
               const apiKey = this._extractApiKey(this._getBody(Capabilities.VOIP_TTS_BODY));
@@ -486,22 +717,28 @@ class BotiumConnectorVoip {
                 apiKey
               });
             }
-            if (Buffer.isBuffer(ttsResponse.data)) {
-              duration = ttsResponse.headers['content-duration'];
+            if (ttsResult && Buffer.isBuffer(ttsResult.buffer)) {
+              duration = parseFloat(ttsResult.duration) || 0;
+              const b64Buffer = ttsResult.buffer.toString('base64');
               const request = JSON.stringify({
                 METHOD: 'sendAudio',
                 PESQ: false,
                 sessionId: this.sessionId,
-                b64_buffer: ttsResponse.data.toString('base64')
+                b64_buffer: b64Buffer
               });
+              if (this._lastBotSaysQueuedAt) {
+                const latencyMs = Date.now() - this._lastBotSaysQueuedAt;
+                const botText = this._lastBotSaysText ? this._lastBotSaysText.substring(0, 120) : '<unknown>';
+                debug$3(`Latency (bot says -> sendAudio/TTS): ${latencyMs} ms (last bot: ${botText})`);
+              }
               msg.attachments.push({
                 name: 'tts.wav',
                 mimeType: 'audio/wav',
-                base64: ttsResponse.data.toString('base64')
+                base64: b64Buffer
               });
               this.ws.send(request);
             } else {
-              reject(new Error(`TTS failed, response is: ${this._getAxiosShortenedOutput(ttsResponse.data)}`));
+              return reject(new Error('TTS failed, response is empty'));
             }
           }
         }
@@ -511,6 +748,11 @@ class BotiumConnectorVoip {
             sessionId: this.sessionId,
             b64_buffer: msg.media[0].buffer.toString('base64')
           });
+          if (this._lastBotSaysQueuedAt) {
+            const latencyMs = Date.now() - this._lastBotSaysQueuedAt;
+            const botText = this._lastBotSaysText ? this._lastBotSaysText.substring(0, 120) : '<unknown>';
+            debug$3(`Latency (bot says -> sendAudio/MEDIA): ${latencyMs} ms (last bot: ${botText})`);
+          }
           this.ws.send(request);
           msg.attachments.push({
             name: msg.media[0].mediaUri,
@@ -535,6 +777,11 @@ class BotiumConnectorVoip {
   }
   async Stop() {
     debug$3(`${this.sessionId} - Stop called`);
+    this.stopCalled = true;
+    if (this.silenceTimeout) {
+      clearTimeout(this.silenceTimeout);
+      this.silenceTimeout = null;
+    }
     if (this.ws && this.ws.readyState !== ws.CLOSED) {
       const request = JSON.stringify({
         METHOD: 'stopCall',
@@ -542,9 +789,12 @@ class BotiumConnectorVoip {
       });
       this.ws.send(request);
       await new Promise(resolve => {
-        setTimeout(resolve, 100000);
-        setInterval(() => {
+        const stopTimeout = setTimeout(resolve, 100000);
+        this._stopWaitInterval = setInterval(() => {
           if (this.end) {
+            clearTimeout(stopTimeout);
+            clearInterval(this._stopWaitInterval);
+            this._stopWaitInterval = null;
             this.wsOpened = false;
             this.ws = null;
             this.end = false;
@@ -562,6 +812,126 @@ class BotiumConnectorVoip {
       this.firstSttInfoReceived = false;
     }
   }
+  _isTtsCacheEnabled() {
+    return this.ttsCacheEnabled && this.ttsCacheMaxEntries > 0;
+  }
+  _touchTtsCache(cacheKey, entry) {
+    if (!this._isTtsCacheEnabled()) return;
+    this.ttsCache.delete(cacheKey);
+    this.ttsCache.set(cacheKey, entry);
+  }
+  _setTtsCacheEntry(cacheKey, entry) {
+    if (!this._isTtsCacheEnabled()) return;
+    this.ttsCache.set(cacheKey, entry);
+    this._trimTtsCache();
+  }
+  _deleteTtsCacheEntry(cacheKey) {
+    if (this.ttsCache) {
+      this.ttsCache.delete(cacheKey);
+    }
+  }
+  _trimTtsCache() {
+    if (!this._isTtsCacheEnabled()) return;
+    while (this.ttsCache.size > this.ttsCacheMaxEntries) {
+      const oldestKey = this.ttsCache.keys().next().value;
+      this.ttsCache.delete(oldestKey);
+    }
+  }
+  _shouldPrefetchTtsText(text) {
+    if (!lodash.isString(text)) return false;
+    if (!text.trim()) return false;
+    return !/\$[A-Za-z]\w*/.test(text);
+  }
+  _maybePrefetchTts(convoStep) {
+    if (!this.ttsPrefetchEnabled || !this._isTtsCacheEnabled()) return;
+    if (!convoStep || convoStep.sender !== 'me') return;
+    if (!this._shouldPrefetchTtsText(convoStep.messageText)) return;
+    const ttsRequest = this._buildTtsRequest(convoStep.messageText);
+    if (!ttsRequest) return;
+    this._getTtsAudio(ttsRequest, convoStep.messageText).catch(err => {
+      debug$3(`TTS prefetch failed - ${this._getAxiosErrOutput(err)}`);
+    });
+  }
+  _buildTtsRequest(text) {
+    if (!this.axiosTtsParams) return null;
+    return {
+      ...this.axiosTtsParams,
+      params: {
+        ...(this.axiosTtsParams.params || {}),
+        text
+      },
+      data: this._getBody(Capabilities.VOIP_TTS_BODY),
+      responseType: 'arraybuffer'
+    };
+  }
+  _getTtsCacheKey(ttsRequest) {
+    const cacheKeyObj = {
+      url: ttsRequest.url,
+      method: ttsRequest.method,
+      params: ttsRequest.params,
+      data: ttsRequest.data
+    };
+    try {
+      return JSON.stringify(cacheKeyObj);
+    } catch (err) {
+      return `${ttsRequest.url}|${ttsRequest.method}|${lodash.get(ttsRequest, 'params.text', '')}`;
+    }
+  }
+  async _fetchTts(ttsRequest, text) {
+    const ttsStart = Date.now();
+    const ttsResponse = await axios(ttsRequest);
+    const ttsMs = Date.now() - ttsStart;
+    const textLen = lodash.isString(text) ? text.length : 0;
+    debug$3(`TTS response ${ttsMs} ms (chars: ${textLen})`);
+    if (!ttsResponse || !Buffer.isBuffer(ttsResponse.data)) {
+      throw new Error(`TTS failed, response is: ${this._getAxiosShortenedOutput(ttsResponse && ttsResponse.data)}`);
+    }
+    const durationHeader = ttsResponse.headers && ttsResponse.headers['content-duration'];
+    return {
+      buffer: ttsResponse.data,
+      duration: durationHeader
+    };
+  }
+  async _getTtsAudio(ttsRequest, text) {
+    if (!ttsRequest) throw new Error('TTS request not configured');
+    const cacheKey = this._getTtsCacheKey(ttsRequest);
+    if (this._isTtsCacheEnabled()) {
+      const cached = this.ttsCache.get(cacheKey);
+      if (cached) {
+        if (cached.state === 'ready') {
+          this._touchTtsCache(cacheKey, cached);
+          debug$3(`TTS cache hit (chars: ${lodash.isString(text) ? text.length : 0})`);
+          return {
+            buffer: cached.buffer,
+            duration: cached.duration
+          };
+        }
+        if (cached.state === 'pending') {
+          return cached.promise;
+        }
+      }
+    }
+    const fetchPromise = this._fetchTts(ttsRequest, text).then(result => {
+      if (this._isTtsCacheEnabled()) {
+        this._setTtsCacheEntry(cacheKey, {
+          state: 'ready',
+          buffer: result.buffer,
+          duration: result.duration
+        });
+      }
+      return result;
+    }).catch(err => {
+      this._deleteTtsCacheEntry(cacheKey);
+      throw err;
+    });
+    if (this._isTtsCacheEnabled()) {
+      this._setTtsCacheEntry(cacheKey, {
+        state: 'pending',
+        promise: fetchPromise
+      });
+    }
+    return fetchPromise;
+  }
   _getParams(capParams) {
     if (this.caps[capParams]) {
       if (lodash.isString(this.caps[capParams])) return JSON.parse(this.caps[capParams]);else return this.caps[capParams];