Danil Sidorov
RU EN DE
Telegram

AI DJ

A private service with a neural network that reproduces light shows

Машинное обучение / Аудио

Live interactive demo stand

An exact copy of the project interface and business logic with demo data. All actions run in real time.

Более 5 площадок повторяют световые шоу

StackPython, Django, PyTorch, Librosa, WebSockets, Channels, Mido

Safe perimeter (NDA): production logic on test data

What to test in the stand:
  1. Запустите аудиотрек и отследите спектрограмму частот в реальном времени
  2. Посмотрите предсказания PyTorch-нейросети для ключевых моментов трека
  3. Проверьте отправку сгенерированных MIDI-команд в световую консоль
http://localhost:8000/
Live Stand
dj_resident

Анализ в реальном времени

Ожидание запуска
Модель: AudioToMidi-v2.4 (BiGRU)
Источник: Virtual Audio Cable (Line 1)
MIDI-порт: LoopBe Internal MIDI
Play
Начать анализ
Используйте эту кнопку для начала стриминга аудио-потока в реальном времени
128 BPM • F1: 0.84
Спектрограмма звука и световая сцена (PyTorch Inference)
Нота: —
Mel: 0 Hz Mel: 4000 Hz Mel: 8000 Hz (fmax)

Журнал событий (WebSocket Stream)

0 событий
Время Обнаруженная нота Точность % MIDI-команда в DJ-софт
Нажмите «Начать анализ», чтобы запустить поток инференса и светомузыку

Turnkey implementation

Need a similar system for your business processes?

I will develop the architecture, fault-tolerant backend, and fast interface. Protected by NDA, clean transferable code, and transparent milestones.

  1. Free timeline and budget estimate within one day
  2. Contract and invoice
  3. Warranty and post-launch support

On weekdays usually within a couple of hours

I will reply personally, no aggressive sales pitch

Task

The client had to trigger effects by hand during a performance. It distracted from the creative work, raised the risk of mistakes in real time and made it hard to focus on the show.

Solution

I built a neural network in PyTorch that predicts the right commands in real time. The network listens to the track and knows when an effect fits and which one, as well as the experienced DJ whose data it was trained on.

What was done

  • Designed an ML pipeline in PyTorch from data collection to real-time inference
  • Trained a neural network to predict the right effect (F1 above 0.8)
  • Added streaming track processing with predictions during playback
  • Integrated the model into the DJ software and removed manual triggering

Result

The neural network now runs the show in semi-automatic mode. The DJ supervises and focuses on the music and the audience, and the routine button pressing is gone.

Technologies

PythonPyTorchDjango

Tell me about your task

Telegram