Kangdi Wang
Curriculum Vitae

Audio research, built on practice.

Perceptual modeling, music-generation development, and audio engineering — from listening evaluation to usable tools.

Download PDF

Kangdi Wang

Audio AI researcher and engineer working at the intersection of neural audio representation, generative music, and professional music production. Contributor to Qwen3.8-Omni-Flash training, focused on agentic audio understanding and REAPER-based arrangement, mixing, and podcast editing. Creator of Rea-Cli, an open-source Python library and CLI for REAPER automation and result verification. Core contributor and second author of Qwen-Music, an industrial-scale music generation system, with hands-on experience in audio quality optimization, music mixing/editing, evaluation, data pipelines, and music agents. First author of εar-VAE (INTERSPEECH 2026) and of εar-VAE2 for high-fidelity stereo music reconstruction with frequency-aware complex-spectral modeling. In parallel, 3+ years of stem/live mixing experience, professional mixing training at Modern Sky, and practical DAW/audio tooling. Interested in generative audio, audio representation, DSP, music understanding, and agentic audio systems.

Email — wangkangdi21@gmail.com GitHub — DDDPG Lab — εar-LAB Scholar — Profile
2026.03 — Present Current

Qwen Omni Foundation Team

Music Generation & Agentic Post-Training Intern
  • Contributed to Qwen-Music decoder development, focusing on audio fidelity and mix/edit capabilities. Core contributor and second-listed author of the technical report.
  • Designed music listening-evaluation protocols and audio-quality annotation standards; developed filtering workflows for training-data curation.
  • Contributed to Qwen3.8-Omni-Flash model training, focusing on agentic audio understanding. Contributor to Qwen3.8-Omni — Towards Native Omni-Modal Agents.
  • Worked on REAPER-based audio editing tasks, including automatic arrangement, mixing, and podcast editing.
2025.05 — 2026.03

Initi AI Ltd

Music Generation Algorithm Intern · Initi AI
  • Contributed to music-generation training pipelines and mix-quality optimization, connecting model development with evaluation of generated audio.
  • Evaluated model outputs and production masters; translated listening observations into feedback for model iteration.
  • Contributed to technical communication for music-generation research.
2024.11 — 2025.05

CereProc Ltd

TTS Development Intern
  • Cleaned Chinese speech corpora and refined text normalization for TTS training data.
  • Contributed to LLM-based TTS development and tuning of the generative vocoder.
  • Debugged the developer SDK and implemented new features.
2023.09 — Present

Xiamen Kazami Technology

Mix Engineer & Audio Developer · Kazami Tech
  • Built venue-specific audio control software on REAPER with Lua and Python — custom widgets and workflow automation for live production.
  • Mixed live-stream events for partners including Bilibili Owl Cup, Bilibili World booth, and related broadcast productions.
  • Designed and optimized on-air audio chains for talent including 小象大鹅, Yuehua Entertainment, VirtuaReal, and NeoNova.
  • 2026.09
    Qwen3.8-Omni: Towards Native Omni-Modal Agents arXiv:2609.25611 · Contributor
  • 2026.08
    Fourier is Frontier: Frequency-Aware Autoencoding for High-Fidelity Music Reconstruction arXiv:2608.19843 · First Author
  • 2026.08
    CineDub: Scaling End-to-End Video Dubbing to Multi-Speaker Dialogues with Coherent Sound Effects ACM Multimedia 2026 · Co-First Author
  • 2026.07
    Qwen-Music Technical Report arXiv:2607.11699 · Core Contributor
  • 2025.11
    DuoTok: Source-Aware Dual-Track Music Tokenization for Vocal-Accompaniment Generation ACM Multimedia 2026 · Co-author
  • 2025.09
    Back to Ear: Perceptually Driven High Fidelity Music Reconstruction INTERSPEECH 2026 · First Author

Publication details →

2021.09 — 2025.06 Graduated

Heriot-Watt University

B.S. in Robotics · HWU
  • Average 83%
  • εar-LAB Co-founded Epsilon Acoustic Revolution Lab — open research initiative for neural audio models, tooling, and community releases.
  • ReaperDoc Created and maintain an interactive reference for REAPER project-file structures and parameters, supporting developers building DAW tooling.
  • SMP-Nx Mixing Contest Contributed translation, English interviews, and editorial writing for the 2024 SMP-Nx mixing competition materials.
  • Elf of Era Idols Project Led production and co-developed a narrative visual novel released on Steam, combining creative production with software delivery.

Audio AI Modeling

  • Perceptual audio autoencoders — εar-VAE / ear-VAE2
  • Music tokenization — DuoTok
  • Music generation and audio rendering — Qwen-Music
  • Multimodal speech and sound generation — CineDub

Signal Processing & Perception

  • Complex STFT, magnitude & phase modeling
  • Perceptual weighting & stereo reconstruction
  • Listening evaluation & audio-quality annotation
  • Speech synthesis & generative vocoders

Audio Production & DAW

  • Python / Lua — REAPER workflow tooling
  • TypeScript / React — audio interfaces
  • Stem mixing & live-stream audio chains
  • REAPER · Pro Tools · Studio One