Delete config.json
Browse files- config.json +0 -90
config.json
DELETED
@@ -1,90 +0,0 @@
|
|
1 |
-
{
|
2 |
-
"model_type": "spark-tts",
|
3 |
-
"architectures": [
|
4 |
-
"SparkTTSModel"
|
5 |
-
],
|
6 |
-
"auto_map": {
|
7 |
-
"AutoConfig": "configuration_spark_tts.SparkTTSConfig",
|
8 |
-
"AutoModel": "modeling_spark_tts.SparkTTSModel",
|
9 |
-
"AutoProcessor": "processing_spark_tts.SparkTTSProcessor"
|
10 |
-
},
|
11 |
-
"processor_class": "processing_spark_tts.SparkTTSProcessor",
|
12 |
-
"custom_pipelines": {
|
13 |
-
"text-to-speech": {
|
14 |
-
"impl": "pipeline_spark_tts.SparkTTSPipeline",
|
15 |
-
"pt": ["AutoModel"]
|
16 |
-
}
|
17 |
-
},
|
18 |
-
"llm_model_name_or_path": "./LLM",
|
19 |
-
"bicodec_model_name_or_path": "./BiCodec",
|
20 |
-
"wav2vec2_model_name_or_path": "./wav2vec2-large-xlsr-53",
|
21 |
-
"sample_rate": 16000,
|
22 |
-
"highpass_cutoff_freq": 40,
|
23 |
-
"latent_hop_length": 320,
|
24 |
-
"ref_segment_duration": 6.0,
|
25 |
-
"volume_normalize": true,
|
26 |
-
"bicodec_config": {
|
27 |
-
"audio_tokenizer": {
|
28 |
-
"mel_params": {
|
29 |
-
"sample_rate": 16000,
|
30 |
-
"n_fft": 1024,
|
31 |
-
"win_length": 640,
|
32 |
-
"hop_length": 320,
|
33 |
-
"mel_fmin": 10,
|
34 |
-
"mel_fmax": null,
|
35 |
-
"num_mels": 128
|
36 |
-
},
|
37 |
-
"encoder": {
|
38 |
-
"input_channels": 1024,
|
39 |
-
"vocos_dim": 384,
|
40 |
-
"vocos_intermediate_dim": 2048,
|
41 |
-
"vocos_num_layers": 12,
|
42 |
-
"out_channels": 1024,
|
43 |
-
"sample_ratios": [1, 1]
|
44 |
-
},
|
45 |
-
"decoder": {
|
46 |
-
"input_channel": 1024,
|
47 |
-
"channels": 1536,
|
48 |
-
"rates": [8, 5, 4, 2],
|
49 |
-
"kernel_sizes": [16, 11, 8, 4]
|
50 |
-
},
|
51 |
-
"quantizer": {
|
52 |
-
"input_dim": 1024,
|
53 |
-
"codebook_size": 8192,
|
54 |
-
"codebook_dim": 8,
|
55 |
-
"commitment": 0.25,
|
56 |
-
"codebook_loss_weight": 2.0,
|
57 |
-
"use_l2_normlize": true,
|
58 |
-
"threshold_ema_dead_code": 0.2
|
59 |
-
},
|
60 |
-
"speaker_encoder": {
|
61 |
-
"input_dim": 128,
|
62 |
-
"out_dim": 1024,
|
63 |
-
"latent_dim": 128,
|
64 |
-
"token_num": 32,
|
65 |
-
"fsq_levels": [4, 4, 4, 4, 4, 4],
|
66 |
-
"fsq_num_quantizers": 1
|
67 |
-
},
|
68 |
-
"prenet": {
|
69 |
-
"input_channels": 1024,
|
70 |
-
"vocos_dim": 384,
|
71 |
-
"vocos_intermediate_dim": 2048,
|
72 |
-
"vocos_num_layers": 12,
|
73 |
-
"out_channels": 1024,
|
74 |
-
"condition_dim": 1024,
|
75 |
-
"sample_ratios": [1, 1],
|
76 |
-
"use_tanh_at_final": false
|
77 |
-
},
|
78 |
-
"postnet": {
|
79 |
-
"input_channels": 1024,
|
80 |
-
"vocos_dim": 384,
|
81 |
-
"vocos_intermediate_dim": 2048,
|
82 |
-
"vocos_num_layers": 6,
|
83 |
-
"out_channels": 1024,
|
84 |
-
"use_tanh_at_final": false
|
85 |
-
}
|
86 |
-
}
|
87 |
-
},
|
88 |
-
"torch_dtype": "bfloat16",
|
89 |
-
"transformers_version": "4.43.1"
|
90 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|