Carnatic Music Rhythm Dataset

CompMusic Carnatic Rhythm Dataset is a rhythm annotated test corpus for automatic rhythm analysis tasks in Carnatic Music. The collection consists of audio excerpts from the CompMusic Carnatic research corpus, manually annotated time aligned markers indicating the progression through the taala cycle, and the associated taala related metadata. A brief description of the dataset is provided below. For a brief overview and audio examples of taalas in Carnatic music, please see


Please cite the following publications if you use the dataset in your work: 

  • Srinivasamurthy, A., Holzapfel, A., Cemgil, A. T., & Serra, X. (2015, October). Particle Filters for Efficient Meter Tracking with Dynamic Bayesian Networks. In Proceedings of the 16th International Society for Music Information Retrieval Conference (ISMIR 2015) (pp. 197–203). Malaga, Spain. (Subset)
  • Srinivasamurthy, A., & Serra, X. (2014, May). A Supervised Approach to Hierarchical Metrical Cycle Tracking from Audio Music Recordings. In Proceedings of the 39th IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2014) (pp. 5237–5241). Florence, Italy. (Full dataset)


The Dataset

Table 1: Description of the dataset
Tāla Beats, Aksharas


Total minutes (hours) Median length of a piece (min) # Annotated beats # Samas Annotated Positions in Tala Cycle
Adi 8, 32 50 252.78 (4.21) 4.85 22793 2882 1,2,3,4,5,6,7,8
Rupaka 3, 12 50 267.45 (4.45) 4.62 22668 7582 1,2,3
Mishra Chapu 7, 14 48 342.13 (5.7) 6.59 31055 7795 1,2,3,4,5,6,7
Khanda Chapu 5, 10 28 134.62 (2.24) 4.41 13111 4387 1,2,3,4,5
Total   176 996.98 (16.62) 5.06 89627 22646  


Audio music content 

The pieces are chosen from the CompMusic Carnatic music collection. The pieces were chosen in four popular taalas of Carnatic music (Table 1), which encompasses a majority of Carnatic music. The pieces were chosen include a mix of vocal and instrumental recordings, new and old recordings, and to span a wide variety of forms. All pieces have a percussion accompaniment, predominantly Mridangam. The excerpts are full length pieces or a part of the full length pieces. There are also several different pieces by the same artist (or release group), and multiple instances of the same composition rendered by different artists. Each piece is uniquely identified using the MBID of the recording. The pieces are stereo, 160 kbps, mp3 files sampled at 44.1 kHz. 


There are several annotations that accompany each excerpt in the dataset. 
Sama and beats: The primary annotations are audio synchronized time-stamps indicating the different metrical positions in the taala cycle. The annotations were created using Sonic Visualizer by tapping to music and manually correcting the taps. Each annotation has a time-stamp and an associated numeric label that indicates the position of the beat marker in the taala cycle. For the metrical positions that are marked in the audio for each taala, please see Figure 1. The marked positions in the taala cycle are shown with numbers, along with the corresponding label used. In each case, the sama (the start of the cycle, analogous to the downbeat) are indicated using the numeral 1. The positions marked in red correspond to the positions in the taala cycle that are displayed through hand gestures in a performance.

1. Adi taala (32 aksharas)


2. Rupaka taala (12 aksharas)   3. Mishra Chapu taala (14 aksharas)   4. Khanda Chapu taala (10 aksharas)

Taala related metadata: For each excerpt, the taala of the piece, edupu (offset of the start of the piece, relative to the sama, measured in aksharas) of the composition, and the kalai (the cycle length scaling factor) are recorded. Each excerpt can be uniquely identified and located with the MBID of the recording, and the relative start and end times of the excerpt within the whole recording. A separate 5 digit taala based unique ID is also provided for each excerpt as a double check. The artist, release, the lead instrument, and the raaga of the piece are additional editorial metadata obtained from the release. A flag indicates if the excerpt is a full piece or only a part of a full piece. There are optional comments on audio quality and annotation specifics. 

Possible uses of the dataset

Possible tasks where the dataset can be used include taala, sama and beat tracking, tempo estimation and tracking, taala recognition, rhythm based segmentation of musical audio, structural segmentation, audio to score/lyrics alignment, and rhythmic pattern discovery. 

Dataset organization

The dataset consists of audio, annotations, an accompanying spreadsheet providing additional metadata. For a detailed description of the organization, please see the README in the dataset. 

Data Subset

A subset of this dataset consisting of 118 two minute excerpts of music is also available. The content in the subset is equaivalent and is separately distributed for a quicker testing of algorithms and approaches. The subset is described below.
Tāla Beats, Aksharas


Total minutes Median length of a piece (min) # Annotated beats # Samas Annotated Positions in Tala Cycle
Adi 8, 32 30 58.87 2 5452 696 1,2,3,4,5,6,7,8
Rupaka 3, 12 30 60 2 5148 1725 1,2,3
Mishra Chapu 7, 14 30 60 2 8992 1299 1,2,3,4,5,6,7
Khanda Chapu 5, 10 28 55.93 2 9133 1840 1,2,3,4,5
Total   118 234.8 2 28725 5560  

Availability of the Dataset

The annotations for the larger dataset and the smaller subset are publicly shared and available to all. The audio is from commercially available releases. It cannot be publicly shared but can be made available on request for non-commercial research purposes. In the future, the dataset would be available for viewing and download through an interface in Dunya ( Please write to us if you wish to use the dataset and need the audio. 


If you have any questions or comments about the dataset, please feel free to write to us. 
Ajay Srinivasamurthy
Music Technology Group
Universitat Pompeu Fabra, 
Barcelona, Spain