File size: 2,698 Bytes
ebf399d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
---
license: apache-2.0
base_model: facebook/wav2vec2-base
tags:
- generated_from_trainer
datasets:
- marsyas/gtzan
metrics:
- accuracy
model-index:
- name: wav2vec2-base-finetuned-gtzan
  results:
  - task:
      name: Audio Classification
      type: audio-classification
    dataset:
      name: GTZAN
      type: marsyas/gtzan
      config: train
      split: train
      args: train
    metrics:
    - name: Accuracy
      type: accuracy
      value: 0.84
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# wav2vec2-base-finetuned-gtzan

This model is a fine-tuned version of [facebook/wav2vec2-base](https://huggingface.co./facebook/wav2vec2-base) on the GTZAN dataset.
It achieves the following results on the evaluation set:
- Loss: 0.8879
- Accuracy: 0.84

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 17

### Training results

| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|:-------------:|:-----:|:----:|:---------------:|:--------:|
| 1.9838        | 1.0   | 113  | 1.8627          | 0.37     |
| 1.6128        | 2.0   | 226  | 1.5998          | 0.48     |
| 1.0259        | 3.0   | 339  | 1.3821          | 0.57     |
| 1.2766        | 4.0   | 452  | 1.1708          | 0.66     |
| 0.6014        | 5.0   | 565  | 0.7257          | 0.77     |
| 0.5815        | 6.0   | 678  | 1.0738          | 0.68     |
| 0.7664        | 7.0   | 791  | 0.7244          | 0.8      |
| 0.2303        | 8.0   | 904  | 0.5838          | 0.84     |
| 0.4829        | 9.0   | 1017 | 0.5741          | 0.87     |
| 0.0859        | 10.0  | 1130 | 0.6199          | 0.83     |
| 0.2983        | 11.0  | 1243 | 0.8117          | 0.84     |
| 0.0642        | 12.0  | 1356 | 0.5938          | 0.88     |
| 0.0688        | 13.0  | 1469 | 0.9978          | 0.84     |
| 0.1542        | 14.0  | 1582 | 0.7437          | 0.85     |
| 0.0117        | 15.0  | 1695 | 0.9100          | 0.84     |
| 0.039         | 16.0  | 1808 | 0.7757          | 0.85     |
| 0.0661        | 17.0  | 1921 | 0.8879          | 0.84     |


### Framework versions

- Transformers 4.31.0
- Pytorch 2.0.1
- Datasets 2.14.0
- Tokenizers 0.13.3