College of Engineering Pune Shivajinagar, Pune-411005. INDIA Dr. S. P. Mahajan Associate Professor, Department of Electronics & Telecommunication E-mail: spm.extc@coep.ac.in; Ph. No. 91-20-25507141 Ph No: +919423005235(M) Name of the candidate: Degree Registered: Title of Thesis: Guide: PhD Thesis Evaluation Report Ms. Surekha Puri Doctor of Philosophy (PhD) in Electronics and Communication Engineering, Savitribai Phule Pune University, Pune A Neural Network Based Automatic Harmonium Music Transcription Prof. S. P. Mahajan General Observations: The candidate has selected a challenging research problem of A Neural Network Based Automatic Harmonium Music Transcription for her PhD thesis which is very much useful for automatic transcription of music for harmonium. The candidate has made a detailed research report organized into seven chapters. The candidate has mentioned 86 references that she has used for her research. The main aim of this thesis is to design the automatic music transcription system for harmonium using neural networks. In order to achieve this aim objectives the candidate has chosen following objectives:1. To the creation of an exhaustive database covering a wide range of harmonium recordings in Indian Classical music. 2. To identification of suitable parameters for Harmonium notes transcription and suggesting neural network-based acoustic model as well as polyphonic model. 3. To maximization of transcription accuracy of Harmonium notes and chords by incorporating neural network based on Music Language Model. 4. To the development of a novel neural network model that will be a combination of acoustic model and Music Language Model for yielding good accuracies. 5. To formulation of a suitable performance evaluation metric for the Harmonium music transcription system. The candidate has conducted an enough experimentation to meet the above objectives. The thesis is well organized and very well written with clarity of thought. The main contributions of this research work can be summarized as follows: 1. Dataset creation of notes, chords, and triads of Harmonium. 2. Identification of particular notes for the Acoustic model(AM)and chord for the Music language model (MLM)for Harmonium music transcription using Convolutional neural network (CNN) and Recurrent Neural Network(RNN). 3. Hybrid model development for harmonium transcription using the Convolutional Recurrent Neural Network(CRNN) model. 4. Conversion of recognized notes and chords into sheet music representation using the lilypond library. 5. Implementation and designing of GUI for AMT are done and which is useful for new learners in the domain of music transcription for the notes and chords transcriptions. Designing of AMT model for monophonic using CNN and RNN approach is well designed for detecting the pitch frequency of a particular note. Various feature extraction techniques such as pitch frequency, MFCC, LPC,CQT and Statistical data for audio notes were implemnted using feed forward neural Networks. The same model is designed for polyphonic (MLM) approach using CNN and RNN . Frequency plot of the detected chords from the corresponding octive is optimised and also sheet music representation is plotted using lilypond library. Hybrid model for harmonium music transcription using the chroma CENS functions from librosa library as input CRNN model is implemented to improve the prediction accuracy of notes and chords. Publications: The candidate has total five publications, including two international conferences and three international journal publications to her credit. Recommendations: Taking into consideration the above observations, it is clear that the thesis submitted by the candidate for her PhD degree has made a good contribution to the knowledge in the broad area of “A Neural Network Based Automatic Harmonium Music Transcription”. The candidate deserves to be complimented for taking up contemporary research topic, conducting the exhaustive experimentation and coming out with useful results. I recommend that the thesis submitted by Ms.Puri Surekha B. be accepted for the award of the Degree of Doctor of Philosophy of Savitribai Phule Pune University, Pune. Pune Date: 29th August 2023 Dr. S. P. Mahajan