{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/sing-symbol-to-instrument-neural-generator","title":"SING: Symbol-to-Instrument Neural Generator","arxiv_id":"1810.09785","date":"2018-10-23","proceeding":"NeurIPS 2018 12","authors":["Alexandre Défossez","Neil Zeghidour","Nicolas Usunier","Léon Bottou","Francis Bach"],"abstract":"Recent progress in deep learning for audio synthesis opens the way to models\nthat directly produce the waveform, shifting away from the traditional paradigm\nof relying on vocoders or MIDI synthesizers for speech or music generation.\nDespite their successes, current state-of-the-art neural audio synthesizers\nsuch as WaveNet and SampleRNN suffer from prohibitive training and inference\ntimes because they are based on autoregressive models that generate audio\nsamples one at a time at a rate of 16kHz. In this work, we study the more\ncomputationally efficient alternative of generating the waveform frame-by-frame\nwith large strides. We present SING, a lightweight neural audio synthesizer for\nthe original task of generating musical notes given desired instrument, pitch\nand velocity. Our model is trained end-to-end to generate notes from nearly\n1000 instruments with a single decoder, thanks to a new loss function that\nminimizes the distances between the log spectrograms of the generated and\ntarget waveforms. On the generalization task of synthesizing notes for pairs of\npitch and instrument not seen during training, SING produces audio with\nsignificantly improved perceptual quality compared to a state-of-the-art\nautoencoder based on WaveNet as measured by a Mean Opinion Score (MOS), and is\nabout 32 times faster for training and 2, 500 times faster for inference.","url_abs":"http://arxiv.org/abs/1810.09785v1","url_pdf":"http://arxiv.org/pdf/1810.09785v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"sing-symbol-to-instrument-neural-generator","repo_url":"https://github.com/facebookresearch/SING","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":"audio-synthesis","task_name":"Audio Synthesis"},{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"music-generation","task_name":"Music Generation"}],"methods":[{"method_slug":"dilated-causal-convolution","method_name":"Dilated Causal Convolution"},{"method_slug":"mixture-of-logistic-distributions","method_name":"Mixture of Logistic Distributions"},{"method_slug":"wavenet","method_name":"WaveNet"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1810.09785","atlas_url":"https://app.syntology.ai/?focus=1810.09785","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}