Gil Weinberg

Naveen Ram, Tanay Gummadi, Rahul Bhethanabotla et al.

Lyric generation is a popular sub-field of natural language generation that has seen growth in recent years. Pop lyrics are of unique interest due to the genre's unique style and content, in addition to the high level of collaboration that goes on behind the scenes in the professional pop songwriting process. In this paper, we present a collaborative line-level lyric generation system that utilizes transfer-learning via the T5 transformer model, which, till date, has not been used to generate pop lyrics. By working and communicating directly with professional songwriters, we develop a model that is able to learn lyrical and stylistic tasks like rhyming, matching line beat requirements, and ending lines with specific target words. Our approach compares favorably to existing methods for multiple datasets and yields positive results from our online studies and interviews with industry songwriters.

Nicholas Farris, Brian Model, Richard Savery et al.

The task of classifying emotions within a musical track has received widespread attention within the Music Information Retrieval (MIR) community. Music emotion recognition has traditionally relied on the use of acoustic features, verbal features, and metadata-based filtering. The role of musical prosody remains under-explored despite several studies demonstrating a strong connection between prosody and emotion. In this study, we restrict the input of traditional machine learning algorithms to the features of musical prosody. Furthermore, our proposed approach builds upon the prior by classifying emotions under an expanded emotional taxonomy, using the Geneva Wheel of Emotion. We utilize a methodology for individual data collection from vocalists, and personal ground truth labeling by the artist themselves. We found that traditional machine learning algorithms when limited to the features of musical prosody (1) achieve high accuracies for a single singer, (2) maintain high accuracy when the dataset is expanded to multiple singers, and (3) achieve high accuracies when trained on a reduced subset of the total features.

Richard Savery, Gil Weinberg

Composing for a film requires developing an understanding of the film, its characters and the film aesthetic choices made by the director. We propose using existing visual analysis systems as a core technology for film music generation. We extract film features including main characters and their emotions to develop a computer understanding of the film's narrative arc. This arc is combined with visually analyzed director aesthetic choices including pacing and levels of movement. Two systems are presented, the first using a robotic film composer and marimbist to generate film scores in real-time performance. The second software-based system builds on the results from the robot film composer to create narrative driven film scores.

Richard Savery, Lisa Zahray, Gil Weinberg

Human collaboration with robotics is dependant on the development of a relationship between human and robot, without which performance and utilization can decrease. Emotion and personality conveyance has been shown to enhance robotic collaborations, with improved human-robot relationships and increased trust. One under-explored way for an artificial agent to convey emotions is through non-linguistic musical prosody. In this work we present a new 4.2 hour dataset of improvised emotional vocal phrases based on the Geneva Emotion Wheel. This dataset has been validated through extensive listening tests and shows promising preliminary results for use in generative systems.

Richard Savery, Lisa Zahray, Gil Weinberg

We present a system for real-time lyrical improvisation between a human and a robot in the style of hip hop. Our system takes vocal input from a human rapper, analyzes the semantic meaning, and generates a response that is rapped back by a robot over a musical groove. Previous work with real-time interactive music systems has largely focused on instrumental output, and vocal interactions with robots have been explored, but not in a musical context. Our generative system includes custom methods for censorship, voice, rhythm, rhyming and a novel deep learning pipeline based on phoneme embeddings. The rap performances are accompanied by synchronized robotic gestures and mouth movements. Key technical challenges that were overcome in the system are developing rhymes, performing with low-latency and dataset censorship. We evaluated several aspects of the system through a survey of videos and sample text output. Analysis of comments showed that the overall perception of the system was positive. The model trained on our hip hop dataset was rated significantly higher than our metal dataset in coherence, rhyme quality, and enjoyment. Participants preferred outputs generated by a given input phrase over outputs generated from unknown keywords, indicating that the system successfully relates its output to its input.

Richard Savery, Lisa Zahray, Gil Weinberg

As robotic arms become prevalent in industry it is crucial to improve levels of trust from human collaborators. Low levels of trust in human-robot interaction can reduce overall performance and prevent full robot utilization. We investigated the potential benefits of using emotional musical prosody to allow the robot to respond emotionally to the user's actions. We tested participants' responses to interacting with a virtual robot arm that acted as a decision agent, helping participants select the next number in a sequence. We compared results from three versions of the application in a between-group experiment, where the robot had different emotional reactions to the user's input depending on whether the user agreed with the robot and whether the user's choice was correct. In all versions, the robot reacted with emotional gestures. One version used prosody-based emotional audio phrases selected from our dataset of singer improvisations, the second version used audio consisting of a single pitch randomly assigned to each emotion, and the final version used no audio, only gestures. Our results showed no significant difference for the percentage of times users from each group agreed with the robot, and no difference between user's agreement with the robot after it made a mistake. However, participants also took a trust survey following the interaction, and we found that the reported trust ratings of the musical prosody group were significantly higher than both the single-pitch and no audio groups.

Ning Yang, Richard Savery, Raghavasimhan Sankaranarayanan et al.

Musical expressivity is an important aspect of musical performance for humans as well as robotic musicians. We present a novel mechatronics-driven implementation of Brushless Direct Current (BLDC) motors in a robotic marimba player, named Shimon, designed to improve speed, dynamic range (loudness), and ultimately perceived musical expressivity in comparison to state-of-the-art robotic percussionist actuators. In an objective test of dynamic range, we find that our implementation provides wider and more consistent dynamic range response in comparison with solenoid-based robotic percussionists. Our implementation also outperforms both solenoid and human marimba players in striking speed. In a subjective listening test measuring musical expressivity, our system performs significantly better than a solenoid-based system and is statistically indistinguishable from human performers.

Richard Savery, Gil Weinberg

As emotion plays a growing role in robotic research it is crucial to develop methods to analyze and compare among the wide range of approaches. To this end we present a survey of 1427 IEEE and ACM publications that include robotics and emotion. This includes broad categorizations of trends in emotion input analysis, robot emotional expression, studies of emotional interaction and models for internal processing. We then focus on 232 papers that present internal processing of emotion, such as using a human's emotion for better interaction or turning environmental stimuli into an emotional drive for robotic path planning. We conducted constant comparison analysis of the 232 papers and arrived at three broad categorization metrics; emotional intelligence, emotional model and implementation, each including two or three subcategories. The subcategories address the algorithm used, emotional mapping, history, the emotional model, emotional categories, the role of emotion, the purpose of emotion and the platform. Our results show a diverse field of study, largely divided by the role of emotion in the system, either for improved interaction, or improved robotic performance. We also present multiple future opportunities for research and describe intrinsic challenges common in all publications.

Richard Savery, Ryan Rose, Gil Weinberg

As human-robot collaboration opportunities continue to expand, trust becomes ever more important for full engagement and utilization of robots. Affective trust, built on emotional relationship and interpersonal bonds is particularly critical as it is more resilient to mistakes and increases the willingness to collaborate. In this paper we present a novel model built on music-driven emotional prosody and gestures that encourages the perception of a robotic identity, designed to avoid uncanny valley. Symbolic musical phrases were generated and tagged with emotional information by human musicians. These phrases controlled a synthesis engine playing back pre-rendered audio samples generated through interpolation of phonemes and electronic instruments. Gestures were also driven by the symbolic phrases, encoding the emotion from the musical phrase to low degree-of-freedom movements. Through a user study we showed that our system was able to accurately portray a range of emotions to the user. We also showed with a significant result that our non-linguistic audio generation achieved an 8% higher mean of average trust than using a state-of-the-art text-to-speech system.

Roozbeh Khodambashi, Gil Weinberg, William Singhose et al.

Supernumerary Robotic Limbs (SRLs) exhibit inherently compliant behavior due to the elasticity present at the intersection of human tissue and the robot. This compliance, can prominently influence the operation of some SRLs, depending on the application. In order to control the residual vibrations of SRLs, we have used an input-shaping method which is a computationally inexpensive approach. The effectiveness of this method in controlling the residual vibrations of a SRL has been proven using robustness analysis. User studies show that reducing the vibrations using input shaping directly increases the user satisfaction and comfort by at least 9%. It is also observed that 36% of the users preferred unshaped commands. We hypothesize that the shaped commands put a higher cognitive load on the user compared to unshaped commands. This shows that when dealing with human-robot interaction, user satisfaction becomes an equally important parameter as traditional performance criteria and should be taken into account while evaluating the success of any vibration-control method.

Mason Bretan, Deepak Gopinath, Philip Mullins et al.

The design and evaluation of a robotic prosthesis for a drummer with a transradial amputation is presented. The principal objective of the prosthesis is to simulate the role fingers play in drumming. This primarily includes controlling the manner in which the drum stick rebounds after initial impact. This is achieved using a DC motor driven by a variable impedance control framework in a shared control system. The user's ability to perform with and control the prosthesis is evaluated using a musical synchronization study. A secondary objective of the prosthesis is to explore the implications of musical expression and human-robotic interaction when a second, completely autonomous, stick is added to the prosthesis. This wearable robotic musician interacts with the user by listening to the music and responding with different rhythms and behaviors. We recount some empirical findings based on the user's experience of performing under such a paradigm.

Mason Bretan, Gil Weinberg, Larry Heck

Several methods exist for a computer to generate music based on data including Markov chains, recurrent neural networks, recombinancy, and grammars. We explore the use of unit selection and concatenation as a means of generating music using a procedure based on ranking, where, we consider a unit to be a variable length number of measures of music. We first examine whether a unit selection method, that is restricted to a finite size unit library, can be sufficient for encompassing a wide spectrum of music. We do this by developing a deep autoencoder that encodes a musical input and reconstructs the input by selecting from the library. We then describe a generative model that combines a deep structured semantic model (DSSM) with an LSTM to predict the next unit, where units consist of four, two, and one measures of music. We evaluate the generative model using objective metrics including mean rank and accuracy and with a subjective listening test in which expert musicians are asked to complete a forced-choiced ranking task. We compare our model to a note-level generative baseline that consists of a stacked LSTM trained to predict forward by one note.