Chris Donahue

Chris Donahue, John Thickstun, Percy Liang · stanford

Despite the central role that melody plays in music perception, it remains an open challenge in music information retrieval to reliably detect the notes of the melody present in an arbitrary music recording. A key challenge in melody transcription is building methods which can handle broad audio containing any number of instrument ensembles and musical styles - existing strategies work well for some melody instruments or styles but not all. To confront this challenge, we leverage representations from Jukebox (Dhariwal et al. 2020), a generative model of broad music audio, thereby improving performance on melody transcription by $20$% relative to conventional spectrogram features. Another obstacle in melody transcription is a lack of training data - we derive a new dataset containing $50$ hours of melody transcriptions from crowdsourced annotations of broad music. The combination of generative pre-training and a new dataset for this task results in $77$% stronger performance on melody transcription relative to the strongest available baseline. By pairing our new melody transcription approach with solutions for beat detection, key estimation, and chord recognition, we build Sheet Sage, a system capable of transcribing human-readable lead sheets directly from music audio. Audio examples can be found at https://chrisdonahue.com/sheetsage and code at https://github.com/chrisdonahue/sheetsage .

John Thickstun, David Hall, Chris Donahue et al. · stanford

We introduce anticipation: a method for constructing a controllable generative model of a temporal point process (the event process) conditioned asynchronously on realizations of a second, correlated process (the control process). We achieve this by interleaving sequences of events and controls, such that controls appear following stopping times in the event sequence. This work is motivated by problems arising in the control of symbolic music generation. We focus on infilling control tasks, whereby the controls are a subset of the events themselves, and conditional generation completes a sequence of events given the fixed control events. We train anticipatory infilling models using the large and diverse Lakh MIDI music dataset. These models match the performance of autoregressive models for prompted music generation, with the additional capability to perform infilling control tasks, including accompaniment. Human evaluators report that an anticipatory model produces accompaniments with similar musicality to even music composed by humans over a 20-second clip.

Chris Donahue, Antoine Caillon, Adam Roberts et al. · stanford

We present SingSong, a system that generates instrumental music to accompany input vocals, potentially offering musicians and non-musicians alike an intuitive new way to create music featuring their own voice. To accomplish this, we build on recent developments in musical source separation and audio generation. Specifically, we apply a state-of-the-art source separation algorithm to a large corpus of music audio to produce aligned pairs of vocals and instrumental sources. Then, we adapt AudioLM (Borsos et al., 2022) -- a state-of-the-art approach for unconditional audio generation -- to be suitable for conditional "audio-to-audio" generation tasks, and train it on the source-separated (vocal, instrumental) pairs. In a pairwise comparison with the same vocal inputs, listeners expressed a significant preference for instrumentals generated by SingSong compared to those from a strong retrieval baseline. Sound examples at https://g.co/magenta/singsong

Aditya Mittal, Ryan Shar, Zichu Wu et al. · cmu

As LLMs are increasingly used as judges in code applications, they should be evaluated in realistic interactive settings that capture partial context and ambiguous intent. We present TRACE (Tool for Rubric Analysis in Code Evaluation), a framework that evaluates LLM judges' ability to predict human preferences and automatically extracts rubric items to reveal systematic biases in how humans and models weigh each item. Across three modalities -- chat-based programming, IDE autocompletion, and instructed code editing -- we use TRACE to measure how well LLM judges align with developer preferences. Among 13 different models, the best judges underperform human annotators by 12-23%. TRACE identifies 35 significant sources of misalignment between humans and judges across interaction modalities, the majority of which correspond to existing software engineering code quality criteria. For example, in chat-based coding, judges are biased towards longer code explanations while humans prefer shorter ones. We find significant misalignment on the majority of existing code quality dimensions, showing alignment gaps between LLM judges and human preference in realistic coding applications.

Yinghao Ma, Anders Øland, Anton Ragni et al.

In recent years, foundation models (FMs) such as large language models (LLMs) and latent diffusion models (LDMs) have profoundly impacted diverse sectors, including music. This comprehensive review examines state-of-the-art (SOTA) pre-trained models and foundation models in music, spanning from representation learning, generative learning and multimodal learning. We first contextualise the significance of music in various industries and trace the evolution of AI in music. By delineating the modalities targeted by foundation models, we discover many of the music representations are underexplored in FM development. Then, emphasis is placed on the lack of versatility of previous methods on diverse music applications, along with the potential of FMs in music understanding, generation and medical application. By comprehensively exploring the details of the model pre-training paradigm, architectural choices, tokenisation, finetuning methodologies and controllability, we emphasise the important topics that should have been well explored, like instruction tuning and in-context learning, scaling law and emergent ability, as well as long-sequence modelling etc. A dedicated section presents insights into music agents, accompanied by a thorough analysis of datasets and evaluations essential for pre-training and downstream tasks. Finally, by underscoring the vital importance of ethical considerations, we advocate that following research on FM for music should focus more on such issues as interpretability, transparency, human responsibility, and copyright issues. The paper offers insights into future challenges and trends on FMs for music, aiming to shape the trajectory of human-AI collaboration in the music realm.

Wayne Chi, Ameet Talwalkar, Chris Donahue

There has been a surge of interest in language model agents that can navigate virtual environments such as the web or desktop. To navigate such environments, agents benefit from information on the various elements (e.g., buttons, text, or images) present. It remains unclear which element attributes have the greatest impact on agent performance, especially in environments that only provide a graphical representation (i.e., pixels). Here we find that the ordering in which elements are presented to the language model is surprisingly impactful--randomizing element ordering in a webpage degrades agent performance comparably to removing all visible text from an agent's state representation. While a webpage provides a hierarchical ordering of elements, there is no such ordering when parsing elements directly from pixels. Moreover, as tasks become more challenging and models more sophisticated, our experiments suggest that the impact of ordering increases. Finding an effective ordering is non-trivial. We investigate the impact of various element ordering methods in web and desktop environments. We find that dimensionality reduction provides a viable ordering for pixel-only environments. We train a UI element detection model to derive elements from pixels and apply our findings to an agent benchmark--OmniACT--where we only have access to pixels. Our method completes more than two times as many tasks on average relative to the previous state-of-the-art.

Wayne Chi, Yixiong Fang, Arnav Yayavaram et al.

Despite rapid progress on coding agents, progress on their multimodal counterparts has lagged behind. A key challenge is the scarcity of evaluation testbeds that combine the complexity of software development with the need for deep multimodal understanding. Game development provides such a testbed as agents must navigate large, dense codebases while manipulating intrinsically multimodal assets such as shaders, sprites, and animations within a visual game scene. We present GameDevBench, the first benchmark for evaluating agents on game development tasks. GameDevBench consists of 132 tasks derived from web and video tutorials. Tasks require significant multimodal understanding and are complex -- the average solution requires over three times the amount of lines of code and file changes compared to prior software development benchmarks. Agents still struggle with game development, with the best agent solving only 54.5% of tasks. We find a strong correlation between perceived task difficulty and multimodal complexity, with success rates dropping from 46.9% on gameplay-oriented tasks to 31.6% on 2D graphics tasks. To improve multimodal capability, we introduce two simple image and video-based feedback mechanisms for agents. Despite their simplicity, these methods consistently improve performance, with the largest change being an increase in Claude Sonnet 4.5's performance from 33.3% to 47.7%. We release GameDevBench publicly to support further research into agentic game development.

Yichen Huang, Zachary Novack, Koichi Saito et al.

Despite significant recent advances in generative acoustic text-to-music (TTM) modeling, robust evaluation of these models lags behind, relying in particular on the popular Fréchet Audio Distance (FAD). In this work, we rigorously study the design space of reference-based divergence metrics for evaluating TTM models through (1) designing four synthetic meta-evaluations to measure sensitivity to particular musical desiderata, and (2) collecting and evaluating on MusicPrefs, the first open-source dataset of human preferences for TTM systems. We find that not only is the standard FAD setup inconsistent on both synthetic and human preference data, but that nearly all existing metrics fail to effectively capture desiderata, and are only weakly correlated with human perception. We propose a new metric, the MAUVE Audio Divergence (MAD), computed on representations from a self-supervised audio embedding model. We find that this metric effectively captures diverse musical desiderata (average rank correlation 0.84 for MAD vs. 0.49 for FAD and also correlates more strongly with MusicPrefs (0.62 vs. 0.14).

Yonghyun Kim, Wayne Chi, Anastasios N. Angelopoulos et al. · gatech

We present Music Arena, an open platform for scalable human preference evaluation of text-to-music (TTM) models. Soliciting human preferences via listening studies is the gold standard for evaluation in TTM, but these studies are expensive to conduct and difficult to compare, as study protocols may differ across systems. Moreover, human preferences might help researchers align their TTM systems or improve automatic evaluation metrics, but an open and renewable source of preferences does not currently exist. We aim to fill these gaps by offering *live* evaluation for TTM. In Music Arena, real-world users input text prompts of their choosing and compare outputs from two TTM systems, and their preferences are used to compile a leaderboard. While Music Arena follows recent evaluation trends in other AI domains, we also design it with key features tailored to music: an LLM-based routing system to navigate the heterogeneous type signatures of TTM systems, and the collection of *detailed* preferences including listening data and natural language feedback. We also propose a rolling data release policy with user privacy guarantees, providing a renewable source of preference data and increasing platform transparency. Through its standardized evaluation protocol, transparent data access policies, and music-specific features, Music Arena not only addresses key challenges in the TTM ecosystem but also demonstrates how live evaluation can be thoughtfully adapted to unique characteristics of specific AI domains. Music Arena is available at: https://music-arena.org . Preference data is available at: https://huggingface.co/music-arena .

Irmak Bukey, Zhepei Wang, Chris Donahue et al.

Music captioning, or the task of generating a natural language description of music, is useful for both music understanding and controllable music generation. Training captioning models, however, typically requires high-quality music caption data which is scarce compared to metadata (e.g., genre, mood, etc.). As a result, it is common to use large language models (LLMs) to synthesize captions from metadata to generate training data for captioning models, though this process imposes a fixed stylization and entangles factual information with natural language style. As a more direct approach, we propose metadata-based captioning. We train a metadata prediction model to infer detailed music metadata from audio and then convert it into expressive captions via pre-trained LLMs at inference time. Compared to a strong end-to-end baseline trained on LLM-generated captions derived from metadata, our method: (1) achieves comparable performance in less training time over end-to-end captioners, (2) offers flexibility to easily change stylization post-training, enabling output captions to be tailored to specific stylistic and quality requirements, and (3) can be prompted with audio and partial metadata to enable powerful metadata imputation or in-filling--a common task for organizing music data.

Karan Goel, Albert Gu, Chris Donahue et al.

Developing architectures suitable for modeling raw audio is a challenging problem due to the high sampling rates of audio waveforms. Standard sequence modeling approaches like RNNs and CNNs have previously been tailored to fit the demands of audio, but the resultant architectures make undesirable computational tradeoffs and struggle to model waveforms effectively. We propose SaShiMi, a new multi-scale architecture for waveform modeling built around the recently introduced S4 model for long sequence modeling. We identify that S4 can be unstable during autoregressive generation, and provide a simple improvement to its parameterization by drawing connections to Hurwitz matrices. SaShiMi yields state-of-the-art performance for unconditional waveform generation in the autoregressive setting. Additionally, SaShiMi improves non-autoregressive generation performance when used as the backbone architecture for a diffusion model. Compared to prior architectures in the autoregressive generation setting, SaShiMi generates piano and speech waveforms which humans find more musical and coherent respectively, e.g. 2x better mean opinion scores than WaveNet on an unconditional speech generation task. On a music generation task, SaShiMi outperforms WaveNet on density estimation and speed at both training and inference even when using 3x fewer parameters. Code can be found at https://github.com/HazyResearch/state-spaces and samples at https://hazyresearch.stanford.edu/sashimi-examples.

Chris Donahue, Shih-Lun Wu, Yewon Kim et al.

We present Hookpad Aria, a generative AI system designed to assist musicians in writing Western pop songs. Our system is seamlessly integrated into Hookpad, a web-based editor designed for the composition of lead sheets: symbolic music scores that describe melody and harmony. Hookpad Aria has numerous generation capabilities designed to assist users in non-sequential composition workflows, including: (1) generating left-to-right continuations of existing material, (2) filling in missing spans in the middle of existing material, and (3) generating harmony from melody and vice versa. Hookpad Aria is also a scalable data flywheel for music co-creation -- since its release in March 2024, Aria has generated 318k suggestions for 3k users who have accepted 74k into their songs. More information about Hookpad Aria is available at https://www.hooktheory.com/hookpad/aria

Irmak Bukey, Michael Feffer, Chris Donahue

We propose an efficient workflow for high-quality offline alignment of in-the-wild performance audio and corresponding sheet music scans (images). Recent work on audio-to-score alignment extends dynamic time warping (DTW) to be theoretically able to handle jumps in sheet music induced by repeat signs-this method requires no human annotations, but we show that it often yields low-quality alignments. As an alternative, we propose a workflow and interface that allows users to quickly annotate jumps (by clicking on repeat signs), requiring a small amount of human supervision but yielding much higher quality alignments on average. Additionally, we refine audio and score feature representations to improve alignment quality by: (1) integrating measure detection into the score feature representation, and (2) using raw onset prediction probabilities from a music transcription model instead of piano roll. We propose an evaluation protocol for audio-to-score alignment that computes the distance between the estimated and ground truth alignment in units of measures. Under this evaluation, we find that our proposed jump annotation workflow and improved feature representations together improve alignment accuracy by 150% relative to prior work (33% to 82%).

Lyria Team, Antoine Caillon, Brian McWilliams et al.

We introduce a new class of generative models for music called live music models that produce a continuous stream of music in real-time with synchronized user control. We release Magenta RealTime, an open-weights live music model that can be steered using text or audio prompts to control acoustic style. On automatic metrics of music quality, Magenta RealTime outperforms other open-weights music generation models, despite using fewer parameters and offering first-of-its-kind live generation capabilities. We also release Lyria RealTime, an API-based model with extended controls, offering access to our most powerful model with wide prompt coverage. These models demonstrate a new paradigm for AI-assisted music creation that emphasizes human-in-the-loop interaction for live music performance.

Yusong Wu, Tim Cooijmans, Kyle Kastner et al.

Jamming requires coordination, anticipation, and collaborative creativity between musicians. Current generative models of music produce expressive output but are not able to generate in an \emph{online} manner, meaning simultaneously with other musicians (human or otherwise). We propose ReaLchords, an online generative model for improvising chord accompaniment to user melody. We start with an online model pretrained by maximum likelihood, and use reinforcement learning to finetune the model for online use. The finetuning objective leverages both a novel reward model that provides feedback on both harmonic and temporal coherency between melody and chord, and a divergence term that implements a novel type of distillation from a teacher model that can see the future melody. Through quantitative experiments and listening tests, we demonstrate that the resulting model adapts well to unfamiliar input and produce fitting accompaniment. ReaLchords opens the door to live jamming, as well as simultaneous co-creation in other modalities.

Xun Zhou, Charlie Ruan, Zihe Zhao et al.

We present the MIDInfinite, a web application capable of generating symbolic music using a large-scale generative AI model locally on commodity hardware. Creating this demo involved porting the Anticipatory Music Transformer, a large language model (LLM) pre-trained on the Lakh MIDI dataset, to the Machine Learning Compilation (MLC) framework. Once the model is ported, MLC facilitates inference on a variety of runtimes including C++, mobile, and the browser. We envision that MLC has the potential to bridge the gap between the landscape of increasingly capable music AI models and technology more familiar to music software developers. As a proof of concept, we build a web application that allows users to generate endless streams of multi-instrumental MIDI in the browser, either from scratch or conditioned on a prompt. On commodity hardware (an M3 Macbook Pro), our demo can generate 51 notes per second, which is faster than real-time playback for 72.9% of generations, and increases to 86.3% with 2 seconds of upfront buffering.

Satvik Dixit, Koichi Saito, Zhi Zhong et al. · cmu

Video-to-audio generation (V2A) is of increasing importance in domains such as film post-production, AR/VR, and sound design, particularly for the creation of Foley sound effects synchronized with on-screen actions. Foley requires generating audio that is both semantically aligned with visible events and temporally aligned with their timing. Yet, there is a mismatch between evaluation and downstream applications due to the absence of a benchmark tailored to Foley-style scenarios. We find that 74% of videos from past evaluation datasets have poor audio-visual correspondence. Moreover, they are dominated by speech and music, domains that lie outside the use case for Foley. To address this gap, we introduce FoleyBench, the first large-scale benchmark explicitly designed for Foley-style V2A evaluation. FoleyBench contains 5,000 (video, ground-truth audio, text caption) triplets, each featuring visible sound sources with audio causally tied to on-screen events. The dataset is built using an automated, scalable pipeline applied to in-the-wild internet videos from YouTube-based and Vimeo-based sources. Compared to past datasets, we show that videos from FoleyBench have stronger coverage of sound categories from a taxonomy specifically designed for Foley sound. Each clip is further labeled with metadata capturing source complexity, UCS/AudioSet category, and video length, enabling fine-grained analysis of model performance and failure modes. We benchmark several state-of-the-art V2A models, evaluating them on audio quality, audio-video alignment, temporal synchronization, and audio-text consistency. Samples are available at: https://gclef-cmu.org/foleybench

Jongmin Jung, Dongmin Kim, Sihun Lee et al.

Music exists in various modalities, such as score images, symbolic scores, MIDI, and audio. Translations between each modality are established as core tasks of music information retrieval, such as automatic music transcription (audio-to-MIDI) and optical music recognition (score image to symbolic score). However, most past work on multimodal translation trains specialized models on individual translation tasks. In this paper, we propose a unified approach, where we train a general-purpose model on many translation tasks simultaneously. Two key factors make this unified approach viable: a new large-scale dataset and the tokenization of each modality. Firstly, we propose a new dataset that consists of more than 1,300 hours of paired audio-score image data collected from YouTube videos, which is an order of magnitude larger than any existing music modal translation datasets. Secondly, our unified tokenization framework discretizes score images, audio, MIDI, and MusicXML into a sequence of tokens, enabling a single encoder-decoder Transformer to tackle multiple cross-modal translation as one coherent sequence-to-sequence task. Experimental results confirm that our unified multitask model improves upon single-task baselines in several key areas, notably reducing the symbol error rate for optical music recognition from 24.58% to a state-of-the-art 13.67%, while similarly substantial improvements are observed across the other translation tasks. Notably, our approach achieves the first successful score-image-conditioned audio generation, marking a significant breakthrough in cross-modal music generation.

Kun Su, Judith Yue Li, Qingqing Huang et al.

Video-to-music generation demands both a temporally localized high-quality listening experience and globally aligned video-acoustic signatures. While recent music generation models excel at the former through advanced audio codecs, the exploration of video-acoustic signatures has been confined to specific visual scenarios. In contrast, our research confronts the challenge of learning globally aligned signatures between video and music directly from paired music and videos, without explicitly modeling domain-specific rhythmic or semantic relationships. We propose V2Meow, a video-to-music generation system capable of producing high-quality music audio for a diverse range of video input types using a multi-stage autoregressive model. Trained on 5k hours of music audio clips paired with video frames mined from in-the-wild music videos, V2Meow is competitive with previous domain-specific models when evaluated in a zero-shot manner. It synthesizes high-fidelity music audio waveforms solely by conditioning on pre-trained general-purpose visual features extracted from video frames, with optional style control via text prompts. Through both qualitative and quantitative evaluations, we demonstrate that our model outperforms various existing music generation systems in terms of visual-audio correspondence and audio quality. Music samples are available at tinyurl.com/v2meow.

Rishi Bommasani, Drew A. Hudson, Ehsan Adeli et al.

AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles(e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities,and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.

Hao-Wen Dong, Chris Donahue, Taylor Berg-Kirkpatrick et al.

Modern keyboards allow a musician to play multiple instruments at the same time by assigning zones -- fixed pitch ranges of the keyboard -- to different instruments. In this paper, we aim to further extend this idea and examine the feasibility of automatic instrumentation -- dynamically assigning instruments to notes in solo music during performance. In addition to the online, real-time-capable setting for performative use cases, automatic instrumentation can also find applications in assistive composing tools in an offline setting. Due to the lack of paired data of original solo music and their full arrangements, we approach automatic instrumentation by learning to separate parts (e.g., voices, instruments and tracks) from their mixture in symbolic multitrack music, assuming that the mixture is to be played on a keyboard. We frame the task of part separation as a sequential multi-class classification problem and adopt machine learning to map sequences of notes into sequences of part labels. To examine the effectiveness of our proposed models, we conduct a comprehensive empirical evaluation over four diverse datasets of different genres and ensembles -- Bach chorales, string quartets, game music and pop music. Our experiments show that the proposed models outperform various baselines. We also demonstrate the potential for our proposed models to produce alternative convincing instrumentations for an existing arrangement by separating its mixture into parts. All source code and audio samples can be found at https://salu133445.github.io/arranger/ .

Rodrigo Castellon, Chris Donahue, Percy Liang

We demonstrate that language models pre-trained on codified (discretely-encoded) music audio learn representations that are useful for downstream MIR tasks. Specifically, we explore representations from Jukebox (Dhariwal et al. 2020): a music generation system containing a language model trained on codified audio from 1M songs. To determine if Jukebox's representations contain useful information for MIR, we use them as input features to train shallow models on several MIR tasks. Relative to representations from conventional MIR models which are pre-trained on tagging, we find that using representations from Jukebox as input features yields 30% stronger performance on average across four MIR tasks: tagging, genre classification, emotion recognition, and key detection. For key detection, we observe that representations from Jukebox are considerably stronger than those from models pre-trained on tagging, suggesting that pre-training via codified audio language modeling may address blind spots in conventional approaches. We interpret the strength of Jukebox's representations as evidence that modeling audio instead of tags provides richer representations for MIR.

Mina Lee, Chris Donahue, Robin Jia et al.

We release a new benchmark for lexical substitution, the task of finding appropriate substitutes for a target word in a context. To assist humans with writing, lexical substitution systems can suggest words that humans cannot easily think of. However, existing benchmarks depend on human recall as the only source of data, and therefore lack coverage of the substitutes that would be most helpful to humans. Furthermore, annotators often provide substitutes of low quality, which are not actually appropriate in the given context. We collect higher-coverage and higher-quality data by framing lexical substitution as a classification problem, guided by the intuition that it is easier for humans to judge the appropriateness of candidate substitutes than conjure them from memory. To this end, we use a context-free thesaurus to produce candidates and rely on human judgement to determine contextual appropriateness. Compared to the previous largest benchmark, our Swords benchmark has 4.1x more substitutes per target word for the same level of quality, and its substitutes are 1.5x more appropriate (based on human judgement) for the same number of substitutes.

Chris Donahue, Mina Lee, Percy Liang

We present a simple approach for text infilling, the task of predicting missing spans of text at any position in a document. While infilling could enable rich functionality especially for writing assistance tools, more attention has been devoted to language modeling---a special case of infilling where text is predicted at the end of a document. In this paper, we aim to extend the capabilities of language models (LMs) to the more general task of infilling. To this end, we train (or fine-tune) off-the-shelf LMs on sequences containing the concatenation of artificially-masked text and the text which was masked. We show that this approach, which we call infilling by language modeling, can enable LMs to infill entire sentences effectively on three different domains: short stories, scientific abstracts, and lyrics. Furthermore, we show that humans have difficulty identifying sentences infilled by our approach as machine-generated in the domain of short stories.

Chris Donahue, Huanru Henry Mao, Yiting Ethan Li et al.

We are interested in the task of generating multi-instrumental music scores. The Transformer architecture has recently shown great promise for the task of piano score generation; here we adapt it to the multi-instrumental setting. Transformers are complex, high-dimensional language models which are capable of capturing long-term structure in sequence data, but require large amounts of data to fit. Their success on piano score generation is partially explained by the large volumes of symbolic data readily available for that domain. We leverage the recently-introduced NES-MDB dataset of four-instrument scores from an early video game sound synthesis chip (the NES), which we find to be well-suited to training with the Transformer architecture. To further improve the performance of our model, we propose a pre-training technique to leverage the information in a large collection of heterogeneous music, namely the Lakh MIDI dataset. Despite differences between the two corpora, we find that this transfer learning procedure improves both quantitative and qualitative performance for our primary task.

Paarth Neekhara, Chris Donahue, Miller Puckette et al.

Recent approaches in text-to-speech (TTS) synthesis employ neural network strategies to vocode perceptually-informed spectrogram representations directly into listenable waveforms. Such vocoding procedures create a computational bottleneck in modern TTS pipelines. We propose an alternative approach which utilizes generative adversarial networks (GANs) to learn mappings from perceptually-informed spectrograms to simple magnitude spectrograms which can be heuristically vocoded. Through a user study, we show that our approach significantly outperforms naïve vocoding strategies while being hundreds of times faster than neural network vocoders used in state-of-the-art TTS systems. We also show that our method can be used to achieve state-of-the-art results in unsupervised synthesis of individual words of speech.

Jesse Engel, Kumar Krishna Agrawal, Shuo Chen et al.

Efficient audio synthesis is an inherently difficult machine learning task, as human perception is sensitive to both global structure and fine-scale waveform coherence. Autoregressive models, such as WaveNet, model local structure at the expense of global latent structure and slow iterative sampling, while Generative Adversarial Networks (GANs), have global latent conditioning and efficient parallel sampling, but struggle to generate locally-coherent audio waveforms. Herein, we demonstrate that GANs can in fact generate high-fidelity and locally-coherent audio by modeling log magnitudes and instantaneous frequencies with sufficient frequency resolution in the spectral domain. Through extensive empirical investigations on the NSynth dataset, we demonstrate that GANs are able to outperform strong WaveNet baselines on automated and human evaluation metrics, and efficiently generate audio several orders of magnitude faster than their autoregressive counterparts.

Chris Donahue, Ian Simon, Sander Dieleman

We present Piano Genie, an intelligent controller which allows non-musicians to improvise on the piano. With Piano Genie, a user performs on a simple interface with eight buttons, and their performance is decoded into the space of plausible piano music in real time. To learn a suitable mapping procedure for this problem, we train recurrent neural network autoencoders with discrete bottlenecks: an encoder learns an appropriate sequence of buttons corresponding to a piano piece, and a decoder learns to map this sequence back to the original piece. During performance, we substitute a user's input for the encoder output, and play the decoder's prediction each time the user presses a button. To improve the intuitiveness of Piano Genie's performance behavior, we impose musically meaningful constraints over the encoder's outputs.

Chris Donahue, Huanru Henry Mao, Julian McAuley

Existing research on music generation focuses on composition, but often ignores the expressive performance characteristics required for plausible renditions of resultant pieces. In this paper, we introduce the Nintendo Entertainment System Music Database (NES-MDB), a large corpus allowing for separate examination of the tasks of composition and performance. NES-MDB contains thousands of multi-instrumental songs composed for playback by the compositionally-constrained NES audio synthesizer. For each song, the dataset contains a musical score for four instrument voices as well as expressive attributes for the dynamics and timbre of each voice. Unlike datasets comprised of General MIDI files, NES-MDB includes all of the information needed to render exact acoustic performances of the original compositions. Alongside the dataset, we provide a tool that renders generated compositions as NES-style audio by emulating the device's audio processor. Additionally, we establish baselines for the tasks of composition, which consists of learning the semantics of composing for the NES synthesizer, and performance, which involves finding a mapping between a composition and realistic expressive attributes.

Chris Donahue, Julian McAuley, Miller Puckette

Audio signals are sampled at high temporal resolutions, and learning to synthesize audio requires capturing structure across a range of timescales. Generative adversarial networks (GANs) have seen wide success at generating images that are both locally and globally coherent, but they have seen little application to audio generation. In this paper we introduce WaveGAN, a first attempt at applying GANs to unsupervised synthesis of raw-waveform audio. WaveGAN is capable of synthesizing one second slices of audio waveforms with global coherence, suitable for sound effect generation. Our experiments demonstrate that, without labels, WaveGAN learns to produce intelligible words when trained on a small-vocabulary speech dataset, and can also synthesize audio from other domains such as drums, bird vocalizations, and piano. We compare WaveGAN to a method which applies GANs designed for image generation on image-like audio feature representations, finding both approaches to be promising.

Chris Donahue, Bo Li, Rohit Prabhavalkar

We investigate the effectiveness of generative adversarial networks (GANs) for speech enhancement, in the context of improving noise robustness of automatic speech recognition (ASR) systems. Prior work demonstrates that GANs can effectively suppress additive noise in raw waveform speech signals, improving perceptual quality metrics; however this technique was not justified in the context of ASR. In this work, we conduct a detailed study to measure the effectiveness of GANs in enhancing speech contaminated by both additive and reverberant noise. Motivated by recent advances in image processing, we propose operating GANs on log-Mel filterbank spectra instead of waveforms, which requires less computation and is more robust to reverberant noise. While GAN enhancement improves the performance of a clean-trained ASR system on noisy speech, it falls short of the performance achieved by conventional multi-style training (MTR). By appending the GAN-enhanced features to the noisy inputs and retraining, we achieve a 7% WER improvement relative to the MTR system.

Chris Donahue, Zachary C. Lipton, Akshay Balsubramani et al.

We propose a new algorithm for training generative adversarial networks that jointly learns latent codes for both identities (e.g. individual humans) and observations (e.g. specific photographs). By fixing the identity portion of the latent codes, we can generate diverse images of the same subject, and by fixing the observation portion, we can traverse the manifold of subjects while maintaining contingent aspects such as lighting and pose. Our algorithm features a pairwise training scheme in which each sample from the generator consists of two images with a common identity code. Corresponding samples from the real dataset consist of two distinct photographs of the same subject. In order to fool the discriminator, the generator must produce pairs that are photorealistic, distinct, and appear to depict the same individual. We augment both the DCGAN and BEGAN approaches with Siamese discriminators to facilitate pairwise training. Experiments with human judges and an off-the-shelf face verification system demonstrate our algorithm's ability to generate convincing, identity-matched photographs.

Chris Donahue, Zachary C. Lipton, Julian McAuley

Dance Dance Revolution (DDR) is a popular rhythm-based video game. Players perform steps on a dance platform in synchronization with music as directed by on-screen step charts. While many step charts are available in standardized packs, players may grow tired of existing charts, or wish to dance to a song for which no chart exists. We introduce the task of learning to choreograph. Given a raw audio track, the goal is to produce a new step chart. This task decomposes naturally into two subtasks: deciding when to place steps and deciding which steps to select. For the step placement task, we combine recurrent and convolutional neural networks to ingest spectrograms of low-level audio features to predict steps, conditioned on chart difficulty. For step selection, we present a conditional LSTM generative model that substantially outperforms n-gram and fixed-window approaches.