EAGER: An Efficient Algorithm for Automated Transcription of Music, Vocalizations, and Arbitrary Sound Recordings

摘要EAGER项目致力于一种新的、健壮的、高效的转录任意声音的技术。 它还以一种新颖且变革性的方式扩展了早期的工作，该工作基于识别美学相似性开发了一个音乐搜索引擎。 音乐声音的自动转录仍然是一个开放的研究领域，也是一个非常重要的领域。 虽然有一些有限的转录技术，但它们缺乏关键的能力，例如无法转录复调作品或难以区分不同来源产生的声音。 此外，频率范围有限，从而排除了大量的音乐作品。 该项目中提出的方法涉及一种创新的音频到音频的转录算法，该算法处理复调作品，捕获谐波，声乐和低音乐器，非常有效，并且可以处理人类制作的音乐作品之外的声音，例如鸟鸣和子/超声波动物发声。

AbstractThis proposed EAGER project focuses on a new, robust and efficient technique to transcribe arbitrary sounds. It also extends in a novel and transformative fashion earlier work that developed a music search engine based on identifying aesthetic similarities. Automated transcription of musical sounds is still an open research area and one of exceptional importance. While some limited transcription techniques are available, they lack critical abilities, such as inability to transcribe polyphonic compositions or difficulties in distinguishing sounds produced by different sources. In addition, the frequency ranges are limited, thereby excluding a vast number of musical works. The approach proposed in this project involves an innovative audio-to-MIDI transcription algorithm, which handles polyphonic compositions, captures harmonic, vocal and percussive instrumentation, is very efficient and works with sounds beyond human produced musical compositions, such as bird songs and sub/ultrasonic animal vocalizations.