Create 12hz_v1_half_config.yaml
Browse files- 12hz_v1_half_config.yaml +32 -0
12hz_v1_half_config.yaml
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
model:
|
| 3 |
+
sample_rate: 16000
|
| 4 |
+
encoder_dim: 32
|
| 5 |
+
encoder_rates: [4,4,5,8,2]
|
| 6 |
+
decoder_rates: [2,8,5,4,4]
|
| 7 |
+
n_codebooks: 24
|
| 8 |
+
quantizer_dropout: 1.0
|
| 9 |
+
codebook_size: 4096
|
| 10 |
+
semantic_codebook_size: 32768
|
| 11 |
+
is_causal: false
|
| 12 |
+
use_similarity_alignment: true
|
| 13 |
+
similarity_threshold: 0.91
|
| 14 |
+
semantic_downsample_factor: 1.33333
|
| 15 |
+
skip_normalize: true
|
| 16 |
+
semantic_model_type: "sensevoice"
|
| 17 |
+
latent_dim: 512
|
| 18 |
+
ssl_dim: 512
|
| 19 |
+
use_bottleneck_transformer: true
|
| 20 |
+
transformer_num_layers: 32
|
| 21 |
+
transformer_dim_feedforward: 2048
|
| 22 |
+
transformer_num_heads: 8
|
| 23 |
+
transformer_causal: false
|
| 24 |
+
transformer_context_frames: 16
|
| 25 |
+
max_tokens_per_group: 8
|
| 26 |
+
semantic_model_path: null
|
| 27 |
+
use_query_token_aggregator: true
|
| 28 |
+
use_dynamic_similarity_threshold: true
|
| 29 |
+
similarity_threshold_lower: 0.8
|
| 30 |
+
similarity_threshold_upper: 1.0
|
| 31 |
+
|
| 32 |
+
|