Conditional Drums Generation Using Compound Word Representations

Music Deep Learning: Deep Learning Methods for Music Signal Processing—A Review of the State-of-the-Art

“…In [197], conditional drum generation is considered, inspired by [166]. A BLSTM encoder receives the condtioning parameter information, and a transformer based decoder with relative global attention generates the drum sequence.…”

“…Apart from hybrid architectures, MDL will be significantly benefited from the use fusion of diverse input modalities. Multiple genres [143], [144], [146], [149], [152] MG (Classical) [150], [153], [ [188], [190], [191], [193], [194] MG (Mandarin Pop) [195] MG (Drums) [197] Melody harmonization [198] This would increase performance, as the conjuction of different modalities can help build connections between different features. For example, in [75] sound signals were extracted from unlabelled video sources.…”

JukeDrummer: Conditional Beat-aware Audio-domain Drum Accompaniment Generation via Transformer VQ-VAE

“…Conditional drum accompaniment generation has been studied in the literature, but only in the symbolic domain [20,21], to the best of our knowledge. Dahale et al [20] used a Transformer encoder to generate an accompanying symbolic drum pattern of 12 bars given a four-track, melodic MIDI passage.…”

“…Dahale et al [20] used a Transformer encoder to generate an accompanying symbolic drum pattern of 12 bars given a four-track, melodic MIDI passage. Makris et al [21] adopted instead a sequence-to-sequence architecture with a bi-directional long short-term memory (BLSTM) encoder extracting information from the melodic input and a Transformer decoder generating the drum track for up to 16 bars in MIDI format. While symbolic-domain music generation has its own challenges, it differs greatly from the audio-domain counterpart studied in this paper, for it is not about generating sounds that can be readily listened to by human.…”

“…• seq2seq+beat (high): using the beat condition embeddings from the high-level beat/downbeat positions as {c m t } instead. 2 A reviewer suggested that we should have compared our model with other accompaniment generation models that operate in the symbolic domain such as [20] and [21], saying that we can use existing audio samples or drum synthesis models to render their output to audio. However, the problem is that such a symbolic-domain accompaniment generation model also requires its input to be a MIDI file rather than audio.…”