metadata

license: mit
inference: false
tags:
  - music

Introduction to our series work

The development log of our Music Audio Pre-training (m-a-p) model family:

17/03/2023: we release two advanced music understanding models, MERT-v1-95M and MERT-v1-330M , trained with new paradigm and dataset. They outperform the previous models and can better generalize to more tasks.
14/03/2023: we retrained the MERT-v0 model with open-source-only music dataset MERT-v0-public
29/12/2022: a music understanding model MERT-v0 trained with MLM paradigm, which performs better at downstream tasks.
29/10/2022: a pre-trained MIR model music2vec trained with BYOL paradigm.

Here is a table for quick model pick-up:

Name	Pre-train Paradigm	Training Data (hour)	Pre-train Context (second)	Model Size	Transformer Layer-Dimension	Feature Rate	Sample Rate	Release Date
MERT-v1-330M	MLM	160K	5	330M	24-1024	75 Hz	24K Hz	17/03/2023
MERT-v1-95M	MLM	20K	5	95M	12-768	75 Hz	24K Hz	17/03/2023
MERT-v0-public	MLM	900	5	95M	12-768	50 Hz	16K Hz	14/03/2023
MERT-v0	MLM	1000	5	95 M	12-768	50 Hz	16K Hz	29/12/2023
music2vec-v1	BYOL	1000	30	95 M	12-768	50 Hz	16K Hz	30/10/2022

Explanation

The m-a-p models share the similar model architecture and the most distinguished difference is the paradigm in used pre-training. Other than that, there are several nuance technical configuration needs to know before using:

Model Size: the number of parameters that would be loaded to memory. Please select the appropriate size fitting your hardware.
Transformer Layer-Dimension: The number of transformer layers and the corresponding feature dimensions can be outputted from our model. This is marked out because features extracted by different layers could have various performance depending on tasks.
Feature Rate: Given a 1-second audio input, the number of features output by the model.
Sample Rate: The frequency of audio that the model is trained with.

Introduction to MERT-v1

Compared to MERT-v0, we introduce multiple new things in the MERT-v1 pre-training:

Change the pseudo labels to 8 codebooks from encodec, which potentially has higher quality and empower our model to support music generation.
MLM prediction with in-batch noise mixture.
Train with higher audio frequency (24K Hz).
Train with more audio data (up to 160 thousands of hours).
More available model sizes 95M and 330M.

More details will be written in our coming-soon paper.

Model Usage

# from transformers import Wav2Vec2Processor
from transformers import Wav2Vec2FeatureExtractor
from transformers import AutoModel
import torch
from torch import nn
import torchaudio.transforms as T
from datasets import load_dataset




commit_hash='b74e8bdecaa1aa58bbd1fd752a7db0695194d9bb'# this is recommended for security reason, the hash might be updated
# loading our model weights
model = AutoModel.from_pretrained("m-a-p/MERT-v1-330M", trust_remote_code=True, revision=commit_hash)
# loading the corresponding preprocessor config
processor = Wav2Vec2FeatureExtractor.from_pretrained("m-a-p/MERT-v1-330M",trust_remote_code=True, revision=commit_hash)

# load demo audio and set processor
dataset = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
dataset = dataset.sort("id")
sampling_rate = dataset.features["audio"].sampling_rate

resample_rate = processor.sampling_rate
# make sure the sample_rate aligned
if resample_rate != sampling_rate:
    print(f'setting rate from {sampling_rate} to {resample_rate}')
    resampler = T.Resample(sampling_rate, resample_rate)
else:
    resampler = None

# audio file is decoded on the fly
if resampler is None:
    input_audio = dataset[0]["audio"]["array"]
else:
  input_audio = resampler(torch.from_numpy(dataset[0]["audio"]["array"]))
  
inputs = processor(input_audio, sampling_rate=resample_rate, return_tensors="pt")
with torch.no_grad():
    outputs = model(**inputs, output_hidden_states=True)

# take a look at the output shape, there are 25 layers of representation
# each layer performs differently in different downstream tasks, you should choose empirically
all_layer_hidden_states = torch.stack(outputs.hidden_states).squeeze()
print(all_layer_hidden_states.shape) # [25 layer, Time steps, 1024 feature_dim]

# for utterance level classification tasks, you can simply reduce the representation in time
time_reduced_hidden_states = all_layer_hidden_states.mean(-2)
print(time_reduced_hidden_states.shape) # [25, 1024]

# you can even use a learnable weighted average representation
aggregator = nn.Conv1d(in_channels=25, out_channels=1, kernel_size=1)
weighted_avg_hidden_states = aggregator(time_reduced_hidden_states.unsqueeze(0)).squeeze()
print(weighted_avg_hidden_states.shape) # [1024]

Citation

@article{li2022large,
  title={Large-Scale Pretrained Model for Self-Supervised Music Audio Representation Learning},
  author={Li, Yizhi and Yuan, Ruibin and Zhang, Ge and Ma, Yinghao and Lin, Chenghua and Chen, Xingran and Ragni, Anton and Yin, Hanzhi and Hu, Zhijie and He, Haoyu and others},
  year={2022}
}