Cog Implementation of MusicGen-Chord

MusicGen Chord is the modified version of Meta's MusicGen Melody model, which can generate music based on audio-based chord conditions or text-based chord conditions.

You can demo this model or learn how to use it with Replicate's API here.

Run with Cog

Cog is an open-source tool that packages machine learning models in a standard, production-ready container. You can deploy your packaged model to your own infrastructure, or to Replicate, where users can interact with it via web interface or API.

Prerequisites

Cog. Follow these instructions to install Cog, or just run:

sudo curl -o /usr/local/bin/cog -L "https://github.com/replicate/cog/releases/latest/download/cog_$(uname -s)_$(uname -m)"
sudo chmod +x /usr/local/bin/cog

Note, to use Cog, you'll also need an installation of Docker.

• GPU machine. You'll need a Linux machine with an NVIDIA GPU attached and the NVIDIA Container Toolkit installed. If you don't already have access to a machine with a GPU, check out our guide to getting a GPU machine.

Step 1. Clone this repository

git clone https://github.com/sakemin/cog-musicgen-chord

Step 2. Run the model

To run the model, you need a local copy of the model's Docker image. You can satisfy this requirement by specifying the image ID in your call to predict like:

cog predict r8.im/sakemin/musicgen-chord@sha256:c940ab4308578237484f90f010b2b3871bf64008e95f26f4d567529ad019a3d6 -i prompt="k pop, cool synthwave, drum and bass with jersey club beats" -i duration=30 -i text_chords="C G A:min F" -i bpm=140 -i time_sig="4/4"

For more information, see the Cog section here

Alternatively, you can build the image yourself, either by running cog build or by letting cog predict trigger the build process implicitly. For example, the following will trigger the build process and then execute prediction:

cog predict -i prompt="k pop, cool synthwave, drum and bass with jersey club beats" -i duration=30 -i text_chords="C G A:min F" -i bpm=140 -i time_sig="4/4"

Note, the first time you run cog predict, model weights and other requisite assets will be downloaded if they're not available locally. This download only needs to be executed once.

Run on replicate

Step 1. Ensure that all assets are available locally

If you haven't already, you should ensure that your model runs locally with cog predict. This will guarantee that all assets are accessible. E.g., run:

cog predict -i prompt="k pop, cool synthwave, drum and bass with jersey club beats" -i duration=30 -i text_chords="C G A:min F" -i bpm=140 -i time_sig="4/4"

Step 2. Create a model on Replicate.

Go to replicate.com/create to create a Replicate model. If you want to keep the model private, make sure to specify "private".

Step 3. Configure the model's hardware

Replicate supports running models on variety of CPU and GPU configurations. For the best performance, you'll want to run this model on an A100 instance.

Click on the "Settings" tab on your model page, scroll down to "GPU hardware", and select "A100". Then click "Save".

Step 4: Push the model to Replicate

Log in to Replicate:

cog login

Push the contents of your current directory to Replicate, using the model name you specified in step 1:

cog push r8.im/username/modelname

Learn more about pushing models to Replicate.

---

Prediction

Prediction Parameters

• prompt (string) : A description of the music you want to generate.
• text_chords (string) : A text based chord progression condition. Single uppercase alphabet character(eg. C) is considered as a major chord. Chord attributes like(maj, min, dim, aug, min6, maj6, min7, minmaj7, maj7, 7, dim7, hdim7, sus2 and sus4) can be added to the root alphabet character after :.(eg. A:min7) Each chord token splitted by SPACE is allocated to a single bar. If more than one chord must be allocated to a single bar, cluster the chords adding with , without any SPACE.(eg. C,C:7 G, E:min A:min) You must choose either only one of audio_chords below or text_chords.
• bpm (number) : BPM condition for the generated output. text_chords will be processed based on this value. This will be appended at the end of prompt.
• time_sig (string) : Time signature value for the generate output. text_chords will be processed based on this value. This will be appended at the end of prompt.
• audio_chord (file) : An audio file that will condition the chord progression. You must choose only one among audio_chords or text_chords above.
• audio_start (integer) : Start time of the audio file to use for chord conditioning.(Default = 0)
• audio_end (integer) : End time of the audio file to use for chord conditioning. If None, will default to the end of the audio clip.
• duration (integer) : Duration of the generated audio in seconds.(Default = 8)
• continuation (boolean) : If True, generated music will continue from audio_chords. If chord conditioning, this is only possible when the chord condition is given with text_chords. If False, generated music will mimic audio_chords's chord.
• multi_band_diffusion (boolean) : If True, the EnCodec tokens will be decoded with MultiBand Diffusion.
• normalization_strategy (string) : Strategy for normalizing audio.(Allowed values : loudness, clip, peak, rms / Default value = loudness)
• top_k (integer) : Reduces sampling to the k most likely tokens.(Default = 250)
• top_p (number) : Reduces sampling to tokens with cumulative probability of p. When set to 0 (default), top_k sampling is used.(Default = 0)
• temperature (number) : Controls the 'conservativeness' of the sampling process. Higher temperature means more diversity.(Default = 1)
• classifier_free_guidance (integer) : Increases the influence of inputs on the output. Higher values produce lower-varience outputs that adhere more closely to inputs.(Default = 3)
• output_format (string) : Output format for generated audio.(Allowed values : wav, mp3 / Default = wav)
• seed (integer) : Seed for random number generator. If None or -1, a random seed will be used.

Text Based Chord Conditioning

Text Chord Condition Format

<progression> ::= <bar> " " <bar>
<bar> ::= <chord> "," <chord>
<chord> ::= <note> ":" <shorthand>
<note> ::= <natural> | <note> <modifier>
<natural> ::= "A" | "B" | "C" | "D" | "E" | "F" | "G"
<modifier> ::= "b" | "#"
<shorthand> ::= "maj" | "min" | "dim" | "aug" | "maj7" | "min7" | "7" | "dim7" | "hdim7" | "minmaj7" | "maj6" | "min6" | "9" | "maj9" | "min9" | "sus4"

• SPACE is used as split token. Each splitted chunk is assigned to a single bar.
  • C G E:min A:min
• When multiple chords must be assigned in a single bar, then append more chords with ,.
  • C G,G:7 E:min,E:min7 A:min
• Chord type can be specified after :.
  • Just using a single uppercase alphabet(eg. C, E) is considered as a major chord.
  • maj, min, dim, aug, min6, maj6, min7, minmaj7, maj7, 7, dim7, hdim7, sus2 and sus4 can be appended with :. - eg. E:dim, B:sus2
• 'sharp' and 'flat' can be specified with # and b.
  • eg. E#:min Db

BPM and Time Signature

• To create chord chroma, bpm and time_sig values must be specified.
  • bpm can be a float value. (eg. 132, 60)
  • The format of time_sig is (int)/(int). (eg. 4/4, 3/4, 6/8, 7/8, 5/4)
• bpm and time_sig values will be automatically concatenated after prompt description value, so you don't need to specify bpm or time signature information in the description for prompt.

Audio Based Chord Conditioning

Audio Chord Conditioning Instruction

• You can also give chord condition with audio_chords.
• With audio_start and audio_end values, you can specify which part of the audio_chords file input will be used as chord condition.
• The chords will be recognized from the audio_chords, using BTC model.

Additional Feature

Continuation

• If continuation is True, then the input audio file given at audio_chords will not be used as audio chord condition. The generated music output will be continued from the given file.
• You can also use audio_start and audio_end values to crop the input audio file.

Infinite Generation

• You can set duration longer than 30 seconds.
• Due to MusicGen's limitation of generating a maximum 30-second audio in one iteration, if the specified duration exceeds 30 seconds, the model will create multiple sequences. It will utilize the latter portion of the output from the previous generation step as the audio prompt (following the same continuation method) for the subsequent generation step.

Multi-Band Diffusion

• Multi-Band Diffusion(MBD) is used for decoding the EnCodec tokens.
• If the tokens are decoded with MBD, than the output audio quality is better.
• Using MBD takes more calculation time, since it has its own prediction sequence.

---

Fine-tuning MusicGen

Assuming you have a local environment configured (i.e. you've completed the steps specified under Run with Cog), you can run training with a command like:

cog train -i dataset_path=@<path-to-your-data> <additional hyperparameters>

Dataset

Audio

• Compressed files in formats like .zip, .tar, .gz, and .tgz are compatible for dataset uploads.
• Single audio files with .mp3, .wav, and .flac formats can also be uploaded.
• Audio files within the dataset must exceed 30 seconds in duration.
• Audio Chunking: Files surpassing 30 seconds will be divided into multiple 30-second chunks.
• Vocal Removal: If drop_vocals is set to True, the vocal tracks in the audio files will be isolated and removed (Default: True).
  • For datasets containing audio without vocals, setting drop_vocals=False reduces data preprocessing time and maintains audio file quality.

Text Description

• If each audio file requires a distinct description, create a .txt file with a single-line description corresponding to each .mp3 or .wav file (e.g., 01_A_Man_Without_Love.mp3 and 01_A_Man_Without_Love.txt).
• For a uniform description across all audio files, set the one_same_description argument to your desired description. In this case, there's no need for individual .txt files.
• Auto Labeling: When auto_labeling is set to True, labels such as 'genre', 'mood', 'theme', 'instrumentation', 'key', and 'bpm' will be generated and added to each audio file in the dataset (Default: True).
  • Available Tags for Labeling

Train Parameters

Train Inputs

• dataset_path: Path = Input("Path to the dataset directory")
• one_same_description: str = Input(description="A description for all audio data", default=None)
• "auto_labeling": bool = Input(description="Generate labels (genre, mood, theme, etc.) for each track using essentia-tensorflow for music information retrieval", default=True)
• "drop_vocals": bool = Input(description="Remove vocal tracks from audio files using Demucs source separation", default=True)
• lr: float = Input(description="Learning rate", default=1)
• epochs: int = Input(description="Number of epochs to train for", default=10)
• updates_per_epoch: int = Input(description="Number of iterations for one epoch", default=100) #If None, iterations per epoch will be set according to dataset/batch size. If a value is provided, the number of iterations per epoch will be set as specified.
• batch_size: int = Input(description="Batch size", default=3)

Default Parameters

• Using epochs=3, updates_per_epoch=100, and lr=1, the fine-tuning process takes approximately 15 minutes.
• For 8 GPU multiprocessing, batch_size must be a multiple of 8. Otherwise, batch_size will be automatically set to the nearest multiple of 8.
• For the chord model, the maximum batch_size is 16 with the specified 8 x Nvidia A40 machine setting.

Example Code with Replicate API

import replicate

training = replicate.trainings.create(
	version="sakemin/musicgen-chord:c940ab4308578237484f90f010b2b3871bf64008e95f26f4d567529ad019a3d6",
  input={
    "dataset_path":"https://your/data/path.zip",
    "one_same_description":"description for your dataset music",
    "epochs":3,
    "updates_per_epoch":100,
  },
  destination="my-name/my-model"
)

print(training)

---

References

• Chord recognition from audio file is performed using BTC model, by Jonggwon Park.
  • Paper : A Bi-Directional Transformer for Musical Chord Recognition
• The auto-labeling feature utilizes effnet-discogs from MTG's essentia.
• 'key' and 'bpm' values are obtained using librosa.
• Vocal dropping is implemented using Meta's demucs.

Licenses

• All code in this repository is licensed under the Apache License 2.0 license.
• The weights in this repository repository are released under the CC-BY-NC 4.0 license as found in the LICENSE_weights file.
• The code in the Audiocraft repository is released under the MIT license (see LICENSE file).
• The weights in the Audiocraft repository are released under the CC-BY-NC 4.0 license (see LICENSE_weights file).