Transcription of piano recordings

Spectral Study with Automatic Formant Extraction to Improve Non-native Pronunciation of English Vowels

“…In this paper, the main idea is based on a previous work [20], tested on a large number of utterances produced by several different speakers; McCandless's discovery was found to be extremely successful. This algorithm is combined with some other ideas already developed by authors [11] in the context of polyphonic piano recordings.…”

“…At the same time, depending on formant position, learners will receive information on mouth opening and tongue positioning according to each vowel sound. Consequently, we are making use of authors' previous research on audio signal processing [11], knowledge on communication channels [12], numerical methods [13], analytical modelling [14] and English applied linguistics [15]. This theoretical framework backs up a useful tool for students of English who want to autonomously improve their pronunciation of English vowels.…”

“…As in [11], our system divides the vowel-segment into temporal slots and, afterwards, a frequency analysis of each slot is done. This temporal segmentation is based on the detection of onsets, so the system is prepared for detecting when a phoneme starts in the recording.…”

“…Secondly, as in [11], we eliminate harmonics of fundamental frequency in each frame except, if we find a peak with higher energy placed in a potential harmonic. The constraint we impose in this step is that the amplitude of a peak placed in the frequency corresponding to the n-th harmonic must be lower than the amplitud of the peak positioned in the frequency corresponding to the n − 1-th harmonic, with n ≤ 1, where, in this notation, the 0-harmonic frequency is the fundamental frequency.…”

“…A sliding window procedure [11] is employed to detect any increases in energy that exceed a certain threshold. This threshold has been selected to characterize the appearance of an onset.…”