01 · AI-ASSISTED SCORE
晴窗 · Qingchuang
An eight-bar instrumental with piano, guitar, bass, strings, and light percussion.
Recording details
0:26 · 80 bpm · C major · Local system-instrument rendering; distinct from neural generation.
RESEARCH · QUESTIONS & EVIDENCE
Research.
A research question becomes useful when its method, data, and limits are clear.
AI + MUSIC · RECENT LOCAL PRACTICE
AI Music Lab
Prompts, musical structure, and three playable outputs.
THE WORKING APPLICATION
The working project.
HEAR THE ACTUAL OUTPUTS
Listen to the outputs.
Download MP3 ↓
02 · MUSICGEN
A piano prompt, generated continuously
A warm piano ballad prompt with acoustic guitar, bass, and brushed drums.
Recording details
0:24 · Seed 2026 · CFG 3 · Temperature 0.85 · Top-k 100 · CPU · 32 kHz.
03 · RIFFUSION
A short piano experiment
Prompt: “gentle piano melody, warm acoustic instrumental, calm and peaceful.”
Recording details
0:05 · Seed 42 · 40 steps · CFG 6 · CPU · 44.1 kHz.
Built: an AI-assisted local interface for generation, experiment records, and audio export.
Solved: CPU fallback when MPS was unavailable; offline score rendering after an initial renderer failed.
Learned: use prompts for exploration and a score for precise musical structure.
Earlier emotion-and-music exploration was supervised by Prof. Zhai at Suzhou University. These recent technical experiments are separate; they do not demonstrate therapeutic effects. MusicGen: Meta, CC-BY-NC 4.0. Riffusion: Seth Forsgren & Hayk Martiros, CreativeML OpenRAIL-M. Score and development were AI-assisted.
DATA · MODELS · INTERPRETATION
Preparing data that others can check.
At AIRBox, Beijing ZeroPlus Technology, I supported flight-recorder data research: source documents, sample indexes, and parameter definitions.
I learned to check units, timing, and sources before comparing data.
MY ROLE
Research & data preparation
I collected and organized reference material, checked sample and parameter information, and recorded uncertainties for team review.
WHAT I PRODUCED
A traceable record
Document lists, sample indexes, comparison notes, and a participation report supported later requirements discussions.
SCOPE
Within a larger engineering team
My responsibility covered early research and documentation. The software screenshots provide team project context; core decoding algorithms and models were outside my role.
RESEARCH SEMINAR · ASTRONOMY & SCIENTIFIC INQUIRY
Learning beyond my own question.
A learning experience connecting the history of astronomical observation with the questions behind my sundial project.
THE CONTEXT
Looking back to ask better questions.
Seminar topics: Chinese astronomy, scientific innovation, and astronomy education.
A connection to my sundial project: making the relationship between sunlight and time visible.
See the sundial project ↗
HOW I WORK
01 · Frame a question02 · Check the evidence03 · Explain the limits04 · Revise the idea