Zema Dataset: A Comprehensive Study of Yaredawi Zema with a Focus on Horologium Chants

TL;DR

The paper addresses the scarcity of computational resources for Ethiopian Orthodox Tewahedo Church chants by introducing a feature-rich Se'atat Zema dataset (Horologium chant) that pairs audio with lyric text and reading-tone labels. It details data collection from a single Horologium source, rigorous audio/text preprocessing, and a multi-step annotation pipeline capturing word-level timings, chanting options, and mode labels, along with quality validation. A preliminary case study using a support vector machine with MFCC features demonstrates the dataset's viability for chanting-mode classification, achieving 69.11% accuracy and highlighting Ezil as the most distinguishable mode. Overall, the dataset (approximately 10.21 hours and 369 clips) enables lyric transcription, lyric-to-audio alignment, and generation tasks, offering a valuable resource for MIR research and the cultural preservation of Yaredawi Zema.

Abstract

Computational music research plays a critical role in advancing music production, distribution, and understanding across various musical styles worldwide. Despite the immense cultural and religious significance, the Ethiopian Orthodox Tewahedo Church (EOTC) chants are relatively underrepresented in computational music research. This paper contributes to this field by introducing a new dataset specifically tailored for analyzing EOTC chants, also known as Yaredawi Zema. This work provides a comprehensive overview of a 10-hour dataset, 369 instances, creation, and curation process, including rigorous quality assurance measures. Our dataset has a detailed word-level temporal boundary and reading tone annotation along with the corresponding chanting mode label of audios. Moreover, we have also identified the chanting options associated with multiple chanting notations in the manuscript by annotating them accordingly. Our goal in making this dataset available to the public 1 is to encourage more research and study of EOTC chants, including lyrics transcription, lyric-to-audio alignment, and music generation tasks. Such research work will advance knowledge and efforts to preserve this distinctive liturgical music, a priceless cultural artifact for the Ethiopian people.

Figures (7)

• Figure 1: The eight basic symbolic notations (neumes) in the EOTC chants. For example: Yizet represents a detached and accented tone (equivalent to staccato), to emphasize a letter or a word. Interested readers on more details are advised to refer to previous articles like Kebede:80bekerie2007st.
• Figure 2: Sireyu combined with low-level milktoch. In the EOTC chant manuscript, lyrics are written in large texts while music notations are written in small texts above the lyrics. The first two rows of notation patterns above the screenshot are extracted from the two sentences, marked with Ge'ez numbers on their left. The first underlined words represent the sireyu while the second and third are immediate example applications of this sireyu. While sireyu has the two letter notations, encircled in green, indicating its "sireyuness," it is represented by its red-circled short-form characters when applied the same melodies at other places.
• Figure 3: Illustration examples of identifying chanting options of repeatedly appearing phrases in audios, each occurrence with different notations. Identification needs some background knowledge and referring to notated manuscripts, applicable in all chanting modes, shown here for Ezil and Ge'ez.
• Figure 4: Illustration of repeated phrases with identical notations in the chant text. Lines 2 to 18 have their first four words omitted due to sharing the same melody as in line 1, while the remaining words vary and are retained. The first four words in line 19 are chanted with the same melody as in line 1, serving as a reference before concluding this pattern of repetition. Similar cases are found throughout our dataset.
• Figure 5: Sample annotation, for Ge'ez YeZema silt of the text shown in Fig. \ref{['fig:optional-notations']}. As there are three chanting options in this mode, we duplicated the two-word phrase three times, though the text appeared once in the book of chant. If we consider the annotation of a recording in Ezil YeZema silt of the same text, the underlined phrase will be duplicated twice as explained in Section \ref{['subsubsec:text-preproc']}. Regarding annotation columns of start4 and end4, for a fourth chanting option in the same mode, we found only two cases and empty values in all others.