Chord Recognition.pdf

This method integrates bass pitch estimation into a hypothesis-search-based chord recognition system. The system uses a probabilistic framework to uniformly integrate bass lines extracted by bass pitch estimation.

Key Components:

1. Bass Pitch Estimation: Bass lines are used as clues for improving chord recognition. Bass sounds are the most predominant tones in the low frequency region and have properties that make them easy to estimate.
2. Probabilistic Framework: A probabilistic framework is used to integrate bass pitch estimation into chord recognition. This framework enables the system to deal with multiple musical elements uniformly and integrate information obtained from statistical analyses of real music.
3. Hypothesis-Search-Based Method: The system uses a hypothesis-search-based method to recognize chord symbols and chord boundaries simultaneously. This method is chosen over HMM-based methods because it expressly solves the mutual dependency problem between chord symbols and chord boundaries.

Evaluation Value Calculation

The evaluation value of the hypothesis is calculated using a weighted sum of three reliabilities:

1. Likelihood of Gaussian Mixture Models: The likelihood of the observed features given the chord and bass pitch.
2. Joint Probability of Chord and Bass Pitch: The joint probability of the chord and bass pitch.
3. Chord Transition N-gram Probability: The probability of the chord transition given the previous chord.

Experimental Results

The system was tested on 150 songs from twelve Beatles albums. The average frame-rate accuracy of the results was 73.4%. The system demonstrates the effectiveness of integrating bass pitch estimation into chord recognition using a probabilistic framework.

Conclusion

The proposed method improves chord recognition by exploiting the interrelationship between chord sequences and bass lines. The probabilistic framework enables the system to deal with multiple musical elements uniformly and integrate information obtained from statistical analyses of real music.