Actually, I anycodings_audio didnÃ¢ÂÂt understand in what way anycodings_audio you are having trouble with the FFT. But I am not sure if that will anycodings_audio do for your accuracy issue. Well, you can try another set of anycodings_audio algorithms for frequency-domain, like anycodings_audio wavelets. anycodings_audio My thesis (Chapter 2) gives a nice anycodings_audio overview. The most anycodings_audio advanced transcribers use automatic anycodings_audio tuning, key information, bass note anycodings_audio information, and information of the anycodings_audio metric position to improve the results. During further anycodings_audio processing, too, they tend to differ anycodings_audio only slightly, though different anycodings_audio time-series smoothing techniques have anycodings_audio been used: hidden Markov models, dynamic anycodings_audio Bayesian networks, support vector anycodings_audio machines (SVMstruct), and conditional anycodings_audio random fields - among others. Most of these approaches use a discrete anycodings_audio Fourier transform (DFT) to create the anycodings_audio initial spectrogram. Try google scholar 'chord anycodings_audio transcription', or 'chord detection', or anycodings_audio 'chord labelling' for advanced research anycodings_audio in this area. ) anycodings_audio are used to tackle this problem. It's very important what you anycodings_audio do afterwards, and often sophisticated anycodings_audio probabilistic models (similar to those anycodings_audio in speech recognition: HMMs, DBNs. It turns out that the way you transform anycodings_audio from the time domain (normal audio) to anycodings_audio the frequency domain (spectral anycodings_audio representation) is only of limited anycodings_audio importance. grouping these pitches over time so as to be able to assign a chord label to a time interval.finding which pitches are present at any time.In fact, anycodings_audio there are two problems (roughly anycodings_audio speaking): Then this is actually a problem that is anycodings_audio slightly removed from recognising the anycodings_audio notes in a piece of audio. If you want a chord representaton of the anycodings_audio song similar to this C G Am F7 F6 C. Good chord anycodings_audio recognition methods could more aptly be anycodings_audio described as 'systems', but usually they anycodings_audio are indeed based on an initial transform anycodings_audio to the frequency domain (most often anycodings_audio DFT). The short answer is that you need much anycodings_audio more than one algorithm.

Chordino chord dictionary