Improving the key extraction performance of a simultaneous local key and chord estimation system

Abstract

In this paper, significant improvements of a previously developed key and chord extraction system are proposed. The major improvement is the introduction of a separate acoustic model, designed to verify local key hypotheses. The conducted experimental evaluation shows that the presented system improves the state of the art in local key estimation. Our experimental study further demonstrates that the chord estimation performance is already quite robust, whereas the key estimation performance still happens to be sensitive to a number of factors. In particular, we present figures that illustrate the significant impact of the embedded musicological model and the duration of the processed excerpt on the key estimation accuracy.

Modeling musicological information as trigrams in a system for simultaneous chord and local key extraction

“…The results of our experiments are summarized in Table 2 and Table 3. They are compared with a bigram modeling system that was previously described in [8]. We use the bigram probabilities of the Kneser-Ney backoff model we constructed here to configure its relative chord model.…”

“…Usually, a finite state automaton is used, with each state either representing a chord [1,2,3,4], a key [5] or a key-chord combination [6,7,8], depending on the desired output. Most of the time, a first order Markov assumption is made such that transitions between states only depend on the previous state.…”

Segmenting music through the joint estimation of keys, chords and structural boundaries

“…Our starting point is the system by Pauwels et al [9] which concurrently estimates keys and chords, but does not estimate structural boundaries. It consists of an HMM in which each state represents a combination of a key and a chord.…”

“…This means that st−1 = O ∨ st = L ⇒ P (kt|st, st−1, kt−1) = δ kt,k t−1 with δ the Kroneckerdelta. For the inter key transitions P (kt|st = O, st−1 = L, kt−1), we reuse the model from [9], based on Lerdahl's theoretical distance [4] between keys.…”

“…However, our proposed method takes a different approach. We present a framework for the joint estimation of structural boundaries, keys and chords that builds upon an existing system for key and chord estimation [9]. It is based on the premise that some chord combinations are more common around structural boundaries, especially when expressed as relative chords in a key, giving a musicologically richer representation that is made possible by the concurrent estimation of keys and chords.…”