핵심 요구사항
- WAV/AIFF/MP3 파일을 입력받는 자율 분석 파이프라인 구축
- BPM·키 및 서브베이스, RMS 밀도, 트랜지언트, 다이내믹 펀치 등 고급 프로덕션 지표 추출
- 드롭·브레이크다운·고에너지 구간 등 구조적 변화 감지 및 매크로/마이크로 에너지 레벨 산출
- 사용자 학습 데이터 없이 동작하는 사전학습/무감독 ML 모델 적용
- 유효한 Rekordbox XML 구조(또는 ID3 태그)로 커스텀 필드(Comments, My Tags) 출력
Required stack
필요 기술
PythonLibrosaEssentiaPyTorchC++JUCEDSPRekordbox XML
Project brief
프로젝트 내용
Looking to hire an experienced Audio DSP / Machine Learning Engineer to build an AI-driven Music Information Retrieval (MIR) engine. The system will extract production-focused audio features from finished tracks and export them directly into Rekordbox-compatible metadata.
Beyond basic tempo and key, the system must leverage Machine Learning models and DSP to automatically calculate deeper sound engineering metrics—sub-bass pressure, full-spectrum RMS density, transient impact, and dynamic punch—and categorize tracks based on acoustic weight and structural energy.
Scope of Work:
Develop an autonomous pipeline/script (Python using Librosa/Essentia/PyTorch, or C++/JUCE) that ingests WAV, AIFF, and MP3 files.
Automatically extract BPM, musical key, and advanced production metrics (sub-bass vs. kick, RMS density, transients, dynamic punch).
Detect structural shifts (drops, breakdowns, high-energy sections) and calculate macro/micro energy levels automatically.
Export analysis results into a fully valid Rekordbox XML structure (or ID3 tags) ready for direct library import.
Acceptance Criteria:
The developer will run the engine on a sample set of audio files. The system must autonomously calculate all production metrics, assign energy levels, and successfully populate custom Rekordbox fields (e.g., Comments, My Tags) upon XML import without requiring manual training data from the user.
Deliverables:
Clean, well-documented code with concise setup instructions so the feature extraction set can be extended in future iterations.
Beyond basic tempo and key, the system must leverage Machine Learning models and DSP to automatically calculate deeper sound engineering metrics—sub-bass pressure, full-spectrum RMS density, transient impact, and dynamic punch—and categorize tracks based on acoustic weight and structural energy.
Scope of Work:
Develop an autonomous pipeline/script (Python using Librosa/Essentia/PyTorch, or C++/JUCE) that ingests WAV, AIFF, and MP3 files.
Automatically extract BPM, musical key, and advanced production metrics (sub-bass vs. kick, RMS density, transients, dynamic punch).
Detect structural shifts (drops, breakdowns, high-energy sections) and calculate macro/micro energy levels automatically.
Export analysis results into a fully valid Rekordbox XML structure (or ID3 tags) ready for direct library import.
Acceptance Criteria:
The developer will run the engine on a sample set of audio files. The system must autonomously calculate all production metrics, assign energy levels, and successfully populate custom Rekordbox fields (e.g., Comments, My Tags) upon XML import without requiring manual training data from the user.
Deliverables:
Clean, well-documented code with concise setup instructions so the feature extraction set can be extended in future iterations.
지원 기회 분석
지원 전에 볼 것
예상 산출물
- WAV/AIFF/MP3를 처리하는 자율 MIR 분석 엔진 스크립트/파이프라인
- Rekordbox 라이브러리에 직접 임포트 가능한 유효한 XML(또는 ID3 태그) 출력
- 깔끔하고 문서화된 코드 및 간결한 셋업 가이드(향후 피처 확장 가능 구조)
매력 포인트
- 스코프와 수용 기준이 비교적 구체적으로 정의됨
- 오디오 DSP/MIR 전문성을 살릴 수 있는 흥미로운 도메인
확인할 점
- 요구 난이도(전문 MIR+ML+DSP)에 비해 예산($250~750)이 현저히 낮음
- '사용자 학습 데이터 없이' 다수의 주관적 지표를 자동 산출하라는 비현실적 기대
- 기간 미명시로 범위 산정이 어려움
Client signal
클라이언트 정보
플랫폼에서 상세 정보를 확인하세요.