Skip to content

replit.md

Latest commit

 

History

History
87 lines (67 loc) · 3.75 KB

replit.md

File metadata and controls

87 lines (67 loc) · 3.75 KB
Raw

ELSA Speech Scorer

A full-stack web app for scoring candidate speech using the ELSA Speech Analyzer API.

Architecture

  • Frontend: React 18 + Vite (port 5000, webview)
  • Backend: Node.js + Express (port 3000, console workflow)
  • Audio processing: FFmpeg (system dependency)

Workflows

  • Start application — Vite dev server on port 5000 (webview)
  • Backend API — Express API server on port 3000

Two Flows

Applicant Flow (/apply) — Public, no login required

  • Candidates submit their own recording (audio or video) via a simple form
  • Fields: Full Name, Email, Recording file, Upload status
  • Submission is stored and visible to admins in the Recruiter Flow
  • Scores are only visible to admins — applicants never see scores

Recruiter Flow (/login → /candidates) — Requires ELSA login

  • Create candidates manually; create meetings per candidate
  • Upload transcript + recording per meeting
  • FFmpeg extracts candidate speech segments from recording using transcript
  • Send to ELSA scoring API; view detailed scores (EPS, CEFR, IELTS, etc.)
  • Applicant submissions appear with an "Applicant" badge and can be scored too

Features

  1. Login — Authenticate with ELSA credentials (Staging or Production)
  2. Candidates — Create/delete; filter by All / Recruiter Flow / Applicant Flow tabs
  3. Meetings — Multiple meetings per candidate; per-meeting upload + scoring
  4. File Upload — Transcript (SRT, VTT, JSON, DOCX, plain text) + recording (any audio/video)
  5. Audio Extraction — FFmpeg extracts candidate speech segments using transcript timestamps; applicant flow uses full audio
  6. ELSA Scoring — Sends extracted audio to ELSA score_audio_plus_elsa API
  7. Score Display — EPS, CEFR, IELTS, Pronunciation, Intonation, Fluency, Grammar, Vocabulary

Project Structure

server/
  index.js                  — Express entry point (serves API + built client)
  routes/
    auth.js                 — POST /api/auth/login
    candidates.js           — CRUD /api/candidates (+ meetings sub-resource)
    scoring.js              — POST /api/scoring/upload/:cid/:mid, /score/:cid/:mid
    apply.js                — POST /api/apply (public applicant upload)
  utils/
    transcriptParser.js     — Parses SRT/VTT/JSON/DOCX/plain text transcripts
    audioExtract.js         — FFmpeg audio segment extraction

client/src/
  App.tsx                   — Router + session context
  pages/
    ApplyPage.tsx           — Public applicant upload form (/apply)
    LoginPage.tsx           — Recruiter login form (/login)
    CandidatesPage.tsx      — Candidate list with flow filter tabs
    CandidateDetailPage.tsx — Candidate info + meeting list
    MeetingDetailPage.tsx   — Per-meeting upload + scoring + score display
  components/
    Layout.tsx              — App shell with header
    ScoreCard.tsx           — Score display with CEFR bars

data/candidates.json        — Persistent candidate/meeting storage
uploads/                    — Uploaded and extracted audio files (gitignored)

Data Model

Candidate: { id, name, email, createdAt, source ('applicant'|undefined), meetings[] } Meeting: { id, name, createdAt, source, transcriptPath, recordingPath, scores }

ELSA API Endpoints

Environment Login Scoring
Staging https://staging.elsanow.co/user/api/v1/account/login https://silencer-stag.elsanow.co/api/v1/score_audio_plus_elsa
Production https://pool.elsanow.io/user/api/v1/account/login https://pool.elsanow.io/api/v1/score_audio_plus_elsa

Deployment

  • Target: VM (always-on; needed for FFmpeg + file persistence)
  • Build: npm run build (installs deps + compiles React)
  • Run: npm start → node server/index.js