owned this note changed 6 years ago
Linked with GitHub

Real-Time Automatic Accompaniment Implementation in Python - 范斯越 Szu-Yueh Fan

歡迎來到 PyCon TW 2019 共筆

Image Not Showing Possible Reasons
  • The image file may be corrupted
  • The server hosting the image is unavailable
  • The image path is incorrect
  • The image format is not supported
Learn More →

共筆入口:https://hackmd.io/@pycontw/2019
手機版請點選上方 按鈕展開議程列表。

PyCon Taiwan 2019 PyNight 活動規劃與 BoF 簽到/建立交流主題
https://hackmd.io/NaI9ymzRQ5urCvwOQ31WAA

從這開始

tags: PyConTW2019

https://github.com/yychen/livespring

  • 即時伴奏(AI)
  • what is automatic accompaniment?
    • 第一input: 樂譜 (包含主奏和伴奏)
    • 第二input: 真人演奏
    • 效果:像是樂團
    • video demo
      • 不管吹奏的快慢,AI仍然可以即時伴奏
  • Roger Dannenbergs
    • 三十年前就被提出來的方法
  • Input
    • 第一種input: MIDI
      • use midi file to present sheet music
      • midi's time period: ticks
      • midi pitch
        • frequency=2^(MIDI pitch-69)*440 Hz
      • MIDI ≈ Sheet Music
        • 不完全相等
        • 但可用
        • 互相轉換會遺失資訊
    • 第二種input: performer's signal
      • analog -> digital
  • System
    • 四個模組:midi module, data module, performance modeule, synthesis module
    • midi module
      • python package:
        • music21
        • pretty_midi
      • separate track
      • adjust bpm
      • track bucketization
    • data module
      • note & signal generator
      • dataset: MusicNet, Bach10v1.1
    • synthesis module
      • pyFluidSynth
    • performance module
      • package sounddevice: 收音
      • feature extraction
        • performer signal -> constant Q Transform(CQT) -> CQT Chroma (12 dimension)
          • 去掉音色的影響
        • package librosa
    • Online processing
  • Research Problem
    • Audio-to-score synchronization
      • offline: 拿到整個樂譜、完整的聲音訊號
        • 應用:interpretation switcher
      • online:
        • 應用:自動翻譜
        • 應用:自動伴奏
  • Decision Model
    • online dynamic time wraping
      • 主奏者訊號 match 樂譜
      • 問題:match的並不好
    • performance hidden Markov model
      • 從決定性的模型轉到機率性的模型
      • emission probability
      • 問題:第一個音下後,伴奏就很急的往前進
    • hierarchical hidden semi-Markov model
      • hidden state transition ~ geometric distribution
      • autoregressive model to predict P
  • Performance Measurement
    • hard limit: 46.43 miliseconds
    • performance average 41.46 < 46.43 miliseconds
  • Github: 18fanfan
Select a repo