Split STT-only and STT+LLM test modes

src/index.ts

@@ -10,11 +10,11 @@ import { OllamaLlmService } from "./services/ollama-llm.js";
 
 const mode = process.argv[2] ?? "test-stt";
 
-async function runSttTest(): Promise<void> {
+async function runSttTest(enableLlm: boolean): Promise<void> {
   const config = loadConfig();
   const logger = new Logger(config.DEBUG ? config.LOG_LEVEL : "error");
   const stt = new FasterWhisperSttService(config, logger);
-  const llm = new OllamaLlmService(config, logger);
+  const llm = enableLlm ? new OllamaLlmService(config, logger) : null;
   let capture = null as ReturnType<typeof spawnLoopbackCapture> | null;
   let shuttingDown: Promise<void> | null = null;
   let receivedChunks = 0;
@@ -69,10 +69,16 @@ async function runSttTest(): Promise<void> {
     void stt.destroy();
   });
 
+  console.log("STT 준비중...");
   await stt.warmup();
   logger.info("STT warmup finished");
-  await llm.warmup();
-  logger.info("LLM warmup finished");
+  console.log("STT 준비 완료");
+  if (llm) {
+    console.log("LLM 준비중...");
+    await llm.warmup();
+    logger.info("LLM warmup finished");
+    console.log("LLM 준비 완료");
+  }
 
   const transcriptionQueue: Array<{ pcm16: Buffer; queuedAt: number; index: number }> = [];
   let transcribing = false;
@@ -108,45 +114,47 @@ async function runSttTest(): Promise<void> {
           console.log(`사용자> ${text}`);
         }
 
-        const assessmentStartedAt = Date.now();
-        const assessment = await llm.assessReplyNeed(text);
-        logger.info("Reply assessment", {
-          index: next.index,
-          should_reply: assessment.shouldReply,
-          likely_needs_lookup: assessment.likelyNeedsLookup,
-          reason: assessment.reason,
-          assessment_ms: Date.now() - assessmentStartedAt,
-        });
+        if (llm) {
+          const assessmentStartedAt = Date.now();
+          const assessment = await llm.assessReplyNeed(text);
+          logger.info("Reply assessment", {
+            index: next.index,
+            should_reply: assessment.shouldReply,
+            likely_needs_lookup: assessment.likelyNeedsLookup,
+            reason: assessment.reason,
+            assessment_ms: Date.now() - assessmentStartedAt,
+          });
 
-        if (!assessment.shouldReply) {
-          if (config.DEBUG) {
-            console.log(`[skip] ${assessment.reason}\n`);
-          }
-          return;
-        }
-
-        const llmStartedAt = Date.now();
-        const reply = await llm.generateReply(text, {
-          onProgress: (message) => {
+          if (!assessment.shouldReply) {
             if (config.DEBUG) {
-              console.log(`[assistant] ${message}`);
-              return;
+              console.log(`[skip] ${assessment.reason}\n`);
             }
-            console.log(`답변> ${message}`);
-          },
-        });
-        logger.info("LLM latency", {
-          index: next.index,
-          llm_ms: Date.now() - llmStartedAt,
-        });
-        logger.info("LLM reply", { index: next.index, text: reply });
-
-        if (config.DEBUG) {
-          if (config.DEBUG_TRANSCRIPTS) {
-            console.log(`[assistant] ${reply}\n`);
+            return;
+          }
+
+          const llmStartedAt = Date.now();
+          const reply = await llm.generateReply(text, {
+            onProgress: (message) => {
+              if (config.DEBUG) {
+                console.log(`[assistant] ${message}`);
+                return;
+              }
+              console.log(`답변> ${message}`);
+            },
+          });
+          logger.info("LLM latency", {
+            index: next.index,
+            llm_ms: Date.now() - llmStartedAt,
+          });
+          logger.info("LLM reply", { index: next.index, text: reply });
+
+          if (config.DEBUG) {
+            if (config.DEBUG_TRANSCRIPTS) {
+              console.log(`[assistant] ${reply}\n`);
+            }
+          } else {
+            console.log(`답변> ${reply}`);
           }
-        } else {
-          console.log(`답변> ${reply}`);
         }
       }
     } catch (error) {
@@ -238,11 +246,14 @@ async function runSttTest(): Promise<void> {
   });
 
   if (config.DEBUG) {
-    console.log("실시간 출력장치 STT 테스트를 시작합니다. Ctrl+C 로 종료합니다.");
+    console.log(enableLlm ? "실시간 출력장치 STT+LLM 테스트를 시작합니다. Ctrl+C 로 종료합니다." : "실시간 출력장치 STT 테스트를 시작합니다. Ctrl+C 로 종료합니다.");
     console.log(`source: ${config.AUDIO_SOURCE ?? "unset"}`);
     console.log(`model: ${config.WHISPER_MODEL}`);
     console.log(`language: ${config.WHISPER_LANGUAGE}`);
     console.log(`beam: ${config.WHISPER_BEAM_SIZE}`);
+    if (enableLlm) {
+      console.log(`llm: ${config.OLLAMA_MODEL}`);
+    }
   }
 
   setInterval(() => {
@@ -325,13 +336,16 @@ async function main(): Promise<void> {
       await printAudioDevices();
       return;
     case "test-stt":
-      await runSttTest();
+      await runSttTest(false);
+      return;
+    case "test-sttllm":
+      await runSttTest(true);
       return;
     case "test-llm":
       await runLlmCli();
       return;
     default:
-      throw new Error(`알 수 없는 실행 모드입니다: ${mode}. 사용 가능: test-stt, test-llm, devices`);
+      throw new Error(`알 수 없는 실행 모드입니다: ${mode}. 사용 가능: test-stt, test-sttllm, test-llm, devices`);
   }
 }
 

src/services/ollama-llm.ts

@@ -376,6 +376,7 @@ export class OllamaLlmService {
         "bun run setup:llm",
         "bun run devices",
         "bun run test:stt",
+        "bun run test:sttllm",
         "bun run test:llm",
       ],
     };