Add audiosep_metadata.json

v1.0.0/audiosep_metadata.json

@@ -0,0 +1,30 @@
+{
+    "model_name": "AudioSep",
+    "architecture": "CLAP TextEncoder + ResUNet30 Separator",
+    "sample_rate": 32000,
+    "audio_channels": 1,
+    "stft_n_fft": 2048,
+    "stft_hop_length": 320,
+    "stft_win_length": 2048,
+    "stft_window": "hann",
+    "stft_center": true,
+    "frequency_bins": 1025,
+    "time_downsample_ratio": 32,
+    "chunk_seconds": 10.0,
+    "chunk_time_frames": 1001,
+    "mask_channels": 3,
+    "text_embedding_dim": 512,
+    "condition_dim": 512,
+    "max_token_length": 512,
+    "text_encoder_model": "AudioSep_TextEncoder",
+    "separator_model": "AudioSep_Separator",
+    "tokenizer_file": "roberta_bpe",
+    "precision": "float32",
+    "total_parameters_millions": 164.0,
+    "text_encoder_parameters_millions": 125.0,
+    "separator_parameters_millions": 39.0,
+    "output_channels": 1,
+    "license": "MIT",
+    "paper": "arXiv:2308.05037",
+    "source": "https://github.com/Audio-AGI/AudioSep"
+}