Feature/#18 음성 면접 기능 stt

.github/workflows/CI_Build_test.yml

@@ -43,4 +43,5 @@ jobs:
           KAKAO_CLIENT_SECRET: ${{ secrets.KAKAO_CLIENT_SECRET }}
           NEXTAUTH_SECRET: ${{ secrets.NEXTAUTH_SECRET }}
           SENTRY_AUTH_TOKEN: ${{ secrets.SENTRY_AUTH_TOKEN }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
         run: pnpm build

src/app/api/ai/stt/route.ts

@@ -0,0 +1,43 @@
+import { ENV } from '@/constants/env-constants';
+import { AI_MESSAGE } from '@/constants/message-constants';
+import { NextRequest, NextResponse } from 'next/server';
+import OpenAI from 'openai';
+
+const openAi = new OpenAI({
+  apiKey: ENV.OPENAI_API_KEY,
+  dangerouslyAllowBrowser: true,
+});
+
+const FORMAT_FORMDATA = {
+  FILE: 'file',
+  MODEL: 'model',
+  LANGUAGE: 'language',
+};
+
+/**
+ * POST 요청 함수
+ */
+export const POST = async (req: NextRequest): Promise<NextResponse> => {
+  const { NOT_FILE, SERVER_ERROR } = AI_MESSAGE.STT;
+  const { FILE, MODEL, LANGUAGE } = FORMAT_FORMDATA;
+  try {
+    const formData = await req.formData();
+    const file = formData.get(FILE) as File;
+    const model = formData.get(MODEL) as string;
+    const language = formData.get(LANGUAGE) as string;
+
+    if (!file) {
+      return NextResponse.json({ message: NOT_FILE }, { status: 400 });
+    }
+
+    const { text } = await openAi.audio.transcriptions.create({
+      file,
+      model,
+      language,
+    });
+
+    return NextResponse.json({ text }, { status: 200 });
+  } catch (error) {
+    return NextResponse.json({ message: SERVER_ERROR }, { status: 503 });
+  }
+};

src/app/api/ai/tts/route.ts

@@ -0,0 +1,42 @@
+import { ENV } from '@/constants/env-constants';
+import { AI_MESSAGE } from '@/constants/message-constants';
+import { NextRequest, NextResponse } from 'next/server';
+import OpenAI from 'openai';
+
+const openAi = new OpenAI({
+  apiKey: ENV.OPENAI_API_KEY,
+  dangerouslyAllowBrowser: true,
+});
+
+/**
+ * POST 요청 함수
+ */
+export const POST = async (req: NextRequest): Promise<NextResponse> => {
+  const { text, model, voice, speed, response_format, instructions } = await req.json();
+  const { REQUEST_FAILURE, SERVER_ERROR } = AI_MESSAGE.TTS;
+  try {
+    const res = await openAi.audio.speech.create({
+      input: text,
+      model,
+      response_format,
+      voice,
+      speed,
+      instructions,
+    });
+
+    if (!res.ok) {
+      const error_message = await res.text();
+      console.error(error_message);
+      return NextResponse.json({ message: REQUEST_FAILURE }, { status: res.status });
+    }
+
+    const arrayBuffer = await res.arrayBuffer();
+    const base64Audio = Buffer.from(arrayBuffer).toString('base64');
+    const audioUrl = `data:audio/${response_format};base64,${base64Audio}`;
+
+    return NextResponse.json({ audioUrl }, { status: 200 });
+  } catch (error) {
+    console.error(error);
+    return NextResponse.json({ message: SERVER_ERROR }, { status: 503 });
+  }
+};

src/constants/api-method-constants.ts

@@ -0,0 +1,7 @@
+export const API_METHOD = {
+  POST: 'POST',
+  GET: 'GET',
+  DELETE: 'DELETE',
+  PUT: 'PUT',
+  PATCH: 'PATCH',
+};

src/constants/env-constants.ts

@@ -14,7 +14,8 @@ export const ENV = {
   NEXTAUTH_URL: process.env.NEXTAUTH_URL,
   NEXTAUTH_SECRET: process.env.NEXTAUTH_SECRET,
 
-  OPENAI_API_KEY: process.env.NEXT_PUBLIC_OPENAI_API_KEY,
+  OPENAI_API_KEY: process.env.OPENAI_API_KEY,
 
-  SENTRY_DSN: process.env.SENTRY_DSN
+  SENTRY_DSN: process.env.SENTRY_DSN,
+  SENTRY_CLIENT_DSN: process.env.NEXT_PUBLIC_SENTRY_DSN,
 };

src/constants/interview-constants.ts

@@ -0,0 +1,4 @@
+export const INTERVIEW_TYPE = {
+  PRESSURE: 'pressure',
+  CALM: 'calm',
+};

src/constants/message-constants.ts

@@ -6,14 +6,25 @@ export const AUTH_MESSAGE = {
     PASSWORD_LENGTH: '비밀번호는 6자 이상이어야 합니다.',
     PASSWORD_SPECIAL_CHAR: '비밀번호는 최소 하나의 특수 문자를 포함해야 합니다.',
     EMAIL_EMPTY_FIELD: '이메일을 입력해주세요.',
-    PASSWORD_EMPTY_FIELD: '비밀번호를 입력해주세요.'
+    PASSWORD_EMPTY_FIELD: '비밀번호를 입력해주세요.',
   },
   RESULT: {
     SIGN_UP_SUCCESS: '회원 가입에 성공했습니다.',
     SIGN_UP_FAILED: '회원 가입에 실패했습니다.',
     SIGN_UP_EXIST_ERROR: '이미 존재하는 이메일입니다.',
     SIGN_UP_EMPTY_FIELD: '모든 값을 입력해주세요.',
     SIGN_IN_SUCCESS: '로그인에 성공했습니다.',
-    SIGN_IN_FAILED: '이메일 혹은 비밀번호를 확인해주세요.'
+    SIGN_IN_FAILED: '이메일 혹은 비밀번호를 확인해주세요.',
+  },
+};
+
+export const AI_MESSAGE = {
+  TTS: {
+    REQUEST_FAILURE: 'TTS 요청 실패',
+    SERVER_ERROR: 'TTS 서버 에러',
+  },
+  STT: {
+    SERVER_ERROR: 'STT 서버 에러',
+    NOT_FILE: '파일이 제공되지 않았습니다.',
   },
 };

src/constants/path-constant.ts

@@ -13,4 +13,8 @@ export const ROUTE_HANDLER_PATH = {
   AUTH: {
     SIGN_UP: '/api/sign-up',
   },
+  AI: {
+    STT: '/api/ai/stt',
+    TTS: '/api/ai/tts',
+  },
 };

src/features/interview/api/client-services.ts

@@ -0,0 +1,102 @@
+import { API_METHOD } from '@/constants/api-method-constants';
+import { ROUTE_HANDLER_PATH } from '@/constants/path-constant';
+import { INTERVIEW_TYPE } from '@/constants/interview-constants';
+import { fetchWithSentry } from '@/utils/fetch-with-sentry';
+
+const { TTS, STT } = ROUTE_HANDLER_PATH.AI;
+const { POST } = API_METHOD;
+
+const TTS_DEFAULT_OPTIONS = {
+  MODEL: 'gpt-4o-mini-tts',
+  FORMAT: 'mp3',
+};
+
+const STT_DEFAULT_OPTIONS = {
+  MODEL: 'gpt-4o-transcribe',
+  FORMAT: 'webm',
+  LANGUAGE: 'ko',
+};
+
+const CALM_OPTIONS = {
+  VOICE: 'ash',
+  SPEED: 1,
+  INSTRUCTION: `Uses a friendly and gentle tone of voice.
+  Rather than challenging the candidate's answers, frequently provides emotional empathy or positive reactions.`,
+};
+
+const PRESSURE_OPTIONS = {
+  VOICE: 'sage',
+  SPEED: 2.5,
+  INSTRUCTION: `Uses a firm and dry tone of voice.
+Avoids showing emotional empathy or positive reactions to the candidate's responses.`,
+};
+
+/**
+ * @function textToSpeech
+ * @param text - 서버로부터 받아온 텍스트 입력
+ * @param model - 고정된 TTS 모델 ('gpt-4o-mini-tts')
+ * @param voice - default = calm('ash'), pressure('sage')
+ * @param speed - default = calm(1), pressure(2.5)
+ * @param response_format  - format 형식 default = 'mp3'
+ * @param instruction  - 목소리 형식 default = calm
+ * @returns {void}
+ */
+
+type TTS_Props = {
+  text: string;
+  type: string;
+};
+
+export const textToSpeech = async ({ text, type }: TTS_Props): Promise<void> => {
+  const { MODEL, FORMAT } = TTS_DEFAULT_OPTIONS;
+  const { PRESSURE } = INTERVIEW_TYPE;
+  const { VOICE, SPEED, INSTRUCTION } = type === PRESSURE ? PRESSURE_OPTIONS : CALM_OPTIONS;
+
+  const res = await fetchWithSentry(TTS, {
+    method: POST,
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      text,
+      model: MODEL,
+      response_format: FORMAT,
+      voice: VOICE,
+      speed: SPEED,
+      instruction: INSTRUCTION,
+    }),
+  });
+
+  const audio = new Audio(res.audioUrl);
+  await audio.play();
+};
+
+/**
+ * @function speechToText
+ * @param blob - 사용자 목소리 파일
+ * @param model - default('whisper-1')
+ * @param format - default('webm')
+ * @param language - default('ko')
+ * @returns {text}
+ */
+
+type STT_Props = {
+  blob: Blob;
+};
+
+export const speechToText = async ({ blob }: STT_Props): Promise<string> => {
+  const { MODEL, FORMAT, LANGUAGE } = STT_DEFAULT_OPTIONS;
+
+  const formData = new FormData();
+
+  const file = new File([blob], 'recording.webm', { type: `audio/${FORMAT}` });
+  formData.append('file', file);
+  formData.append('model', MODEL);
+  formData.append('language', LANGUAGE);
+  formData.append('format', FORMAT);
+
+  const res = await fetchWithSentry(STT, {
+    method: POST,
+    body: formData,
+  });
+
+  return res.text;
+};