Hugging Face's logo

bosonai
/
higgs-audio-v2-tokenizer

Feature Extraction
Transformers
Safetensors
higgs_audio_v2_tokenizer
Model card Files
xet
 Community
7

trfms-support

#4
by eustlb HF Staff - opened
base: refs/heads/main
←
from: refs/pr/4
Discussion Files changed
+149
-8
Files changed (4) hide show
  1. README.md +12 -0
  2. config.json +125 -8
  3. model.safetensors +3 -0
  4. preprocessor_config.json +9 -0
README.md CHANGED
@@ -1,5 +1,6 @@
1
  ---
2
  license: other
 
3
  ---
4
 
5
  # Higgs Audio Tokenizer
@@ -18,6 +19,17 @@ We introduce a new discretized audio tokenizer that runs at just **25 frames per
18
  <img src="./higgs_audio_tokenizer_architecture.png" alt="Architecture diagram of the Higgs Audio Tokenizer" width=900>
19
  </p>
20
 
 
 
 
 
 
 
 
 
 
 
 
21
  ## Basics of Audio Quantization
22
 
23
  An audio signal sampled at \\(f_s\\) Hz is first split into frames by an encoder with hop size \\(M\\), giving a frame rate \\(f_r = \frac{f_s}{M}\quad\text{(frames/s)}.\\)
 
1
  ---
2
  license: other
3
+ library_name: transformers
4
  ---
5
 
6
  # Higgs Audio Tokenizer
 
19
  <img src="./higgs_audio_tokenizer_architecture.png" alt="Architecture diagram of the Higgs Audio Tokenizer" width=900>
20
  </p>
21
 
22
+ ## Usage
23
+
24
+ ### Transformers 🤗
25
+
26
+ Higgs Audio V2 Tokenizer is supported natively in `transformers`: [see the doc](https://huggingface.co/docs/transformers/en/model_doc/higgs_audio_v2_tokenizer).
27
+
28
+ ```bash
29
+ uv pip install "transformers>=5.3.0"
30
+ ```
31
+
32
+
33
  ## Basics of Audio Quantization
34
 
35
  An audio signal sampled at \\(f_s\\) Hz is first split into frames by an encoder with hop size \\(M\\), giving a frame rate \\(f_r = \frac{f_s}{M}\quad\text{(frames/s)}.\\)
config.json CHANGED
@@ -1,12 +1,129 @@
1
  {
2
- "n_filters": 32,
3
- "D": 256,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4
  "codebook_dim": 64,
5
- "target_bandwidths": [0.5, 1, 1.5, 2, 4],
6
- "ratios": [8, 5, 4, 2, 3],
 
 
 
 
7
  "sample_rate": 24000,
8
- "bins": 1024,
9
- "n_q": 8,
10
- "semantic_techer": "hubert_base_general"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  }
12
-
 
1
  {
2
+ "acoustic_model_config": {
3
+ "codebook_dim": 8,
4
+ "codebook_loss_weight": 1.0,
5
+ "codebook_size": 1024,
6
+ "commitment_loss_weight": 0.25,
7
+ "decoder_hidden_size": 1024,
8
+ "downsampling_ratios": [
9
+ 8,
10
+ 5,
11
+ 4,
12
+ 2,
13
+ 3
14
+ ],
15
+ "encoder_hidden_size": 64,
16
+ "hidden_size": 256,
17
+ "hop_length": 960,
18
+ "model_type": "dac",
19
+ "n_codebooks": 9,
20
+ "quantizer_dropout": 0,
21
+ "sampling_rate": 16000,
22
+ "upsampling_ratios": [
23
+ 8,
24
+ 5,
25
+ 4,
26
+ 2,
27
+ 3
28
+ ]
29
+ },
30
+ "architectures": [
31
+ "HiggsAudioV2TokenizerModel"
32
+ ],
33
+ "block_dilations": [
34
+ 1,
35
+ 1
36
+ ],
37
+ "channel_ratios": [
38
+ 1,
39
+ 1
40
+ ],
41
  "codebook_dim": 64,
42
+ "codebook_size": 1024,
43
+ "downsample_factor": 320,
44
+ "dtype": "float32",
45
+ "initializer_range": 0.02,
46
+ "kernel_size": 3,
47
+ "model_type": "higgs_audio_v2_tokenizer",
48
  "sample_rate": 24000,
49
+ "semantic_model_config": {
50
+ "activation_dropout": 0.1,
51
+ "apply_spec_augment": true,
52
+ "attention_dropout": 0.1,
53
+ "bos_token_id": 1,
54
+ "classifier_proj_size": 256,
55
+ "conv_bias": false,
56
+ "conv_dim": [
57
+ 512,
58
+ 512,
59
+ 512,
60
+ 512,
61
+ 512,
62
+ 512,
63
+ 512
64
+ ],
65
+ "conv_kernel": [
66
+ 10,
67
+ 3,
68
+ 3,
69
+ 3,
70
+ 3,
71
+ 2,
72
+ 2
73
+ ],
74
+ "conv_pos_batch_norm": false,
75
+ "conv_stride": [
76
+ 5,
77
+ 2,
78
+ 2,
79
+ 2,
80
+ 2,
81
+ 2,
82
+ 2
83
+ ],
84
+ "ctc_loss_reduction": "sum",
85
+ "ctc_zero_infinity": false,
86
+ "do_stable_layer_norm": false,
87
+ "eos_token_id": 2,
88
+ "feat_extract_activation": "gelu",
89
+ "feat_extract_norm": "group",
90
+ "feat_proj_dropout": 0.0,
91
+ "feat_proj_layer_norm": true,
92
+ "final_dropout": 0.1,
93
+ "hidden_act": "gelu",
94
+ "hidden_dropout": 0.1,
95
+ "hidden_size": 768,
96
+ "initializer_range": 0.02,
97
+ "intermediate_size": 3072,
98
+ "layer_norm_eps": 1e-05,
99
+ "layerdrop": 0.1,
100
+ "mask_feature_length": 10,
101
+ "mask_feature_min_masks": 0,
102
+ "mask_feature_prob": 0.0,
103
+ "mask_time_length": 10,
104
+ "mask_time_min_masks": 2,
105
+ "mask_time_prob": 0.0,
106
+ "model_type": "hubert",
107
+ "num_attention_heads": 12,
108
+ "num_conv_pos_embedding_groups": 16,
109
+ "num_conv_pos_embeddings": 128,
110
+ "num_feat_extract_layers": 7,
111
+ "num_hidden_layers": 12,
112
+ "pad_token_id": 0,
113
+ "use_weighted_layer_sum": false,
114
+ "vocab_size": 32
115
+ },
116
+ "semantic_sample_rate": 16000,
117
+ "strides": [
118
+ 1,
119
+ 1
120
+ ],
121
+ "target_bandwidths": [
122
+ 0.5,
123
+ 1,
124
+ 1.5,
125
+ 2
126
+ ],
127
+ "transformers_version": "5.3.0.dev0",
128
+ "unit_kernel_size": 3
129
  }
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9c896b019ec1a4fc950218e32797c30870d36ce1cd76f45b35de46495764a204
3
+ size 11542613696
preprocessor_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "feature_extractor_type": "DacFeatureExtractor",
3
+ "feature_size": 1,
4
+ "hop_length": 960,
5
+ "padding_side": "right",
6
+ "padding_value": 0.0,
7
+ "return_attention_mask": true,
8
+ "sampling_rate": 24000
9
+ }