Update audioldm/utils.py

audioldm/utils.py

@@ -69,5 +69,106 @@ def instantiate_from_config(config):
         raise KeyError("Expected key `target` to instantiate.")
     return get_obj_from_str(config["target"])(**config.get("params", dict()))
 
-def default_audioldm_config():
-    return {'wave_file_save_path': './output', 'id': {'version': 'v1', 'name': 'default', 'root': '/mnt/fast/nobackup/users/hl01486/projects/general_audio_generation/AudioLDM-python/config/default/latent_diffusion.yaml'}, 'model': {'device': 'cuda', 'reload_from_ckpt': '/mnt/fast/nobackup/scratch4weeks/hl01486/exps/audio_generation/stablediffusion/LDM/audioverse/2023_01_14_full_F4_B_spatial_v2_v1/checkpoints/last.ckpt', 'target': 'audioldm.pipline.LatentDiffusion', 'params': {'base_learning_rate': 5e-06, 'linear_start': 0.0015, 'linear_end': 0.0195, 'num_timesteps_cond': 1, 'log_every_t': 200, 'timesteps': 1000, 'first_stage_key': 'fbank', 'cond_stage_key': 'waveform', 'latent_t_size': 256, 'latent_f_size': 16, 'channels': 8, 'cond_stage_trainable': True, 'conditioning_key': 'film', 'monitor': 'val/loss_simple_ema', 'scale_by_std': True, 'unet_config': {'target': 'audioldm.latent_diffusion.openaimodel.UNetModel', 'params': {'image_size': 64, 'extra_film_condition_dim': 512, 'extra_film_use_concat': True, 'in_channels': 8, 'out_channels': 8, 'model_channels': 128, 'attention_resolutions': [8, 4, 2], 'num_res_blocks': 2, 'channel_mult': [1, 2, 3, 5], 'num_head_channels': 32, 'use_spatial_transformer': True}}, 'first_stage_config': {'base_learning_rate': 4.5e-05, 'target': 'audioldm.variational_autoencoder.autoencoder.AutoencoderKL', 'params': {'monitor': 'val/rec_loss', 'image_key': 'fbank', 'subband': 1, 'embed_dim': 8, 'time_shuffle': 1, 'ddconfig': {'double_z': True, 'z_channels': 8, 'resolution': 256, 'downsample_time': False, 'in_channels': 1, 'out_ch': 1, 'ch': 128, 'ch_mult': [1, 2, 4], 'num_res_blocks': 2, 'attn_resolutions': [], 'dropout': 0.0}}}, 'cond_stage_config': {'target': 'audioldm.clap.encoders.CLAPAudioEmbeddingClassifierFreev2', 'params': {'key': 'waveform', 'sampling_rate': 16000, 'embed_mode': 'audio', 'unconditional_prob': 0.1}}}}}
+def default_audioldm_config(model_name="audioldm-s-full"):    
+    basic_config = {
+        "wave_file_save_path": "./output",
+        "id": {
+            "version": "v1",
+            "name": "default",
+            "root": "/mnt/fast/nobackup/users/hl01486/projects/general_audio_generation/AudioLDM-python/config/default/latent_diffusion.yaml",
+        },
+        "preprocessing": {
+            "audio": {"sampling_rate": 16000, "max_wav_value": 32768},
+            "stft": {"filter_length": 1024, "hop_length": 160, "win_length": 1024},
+            "mel": {
+                "n_mel_channels": 64,
+                "mel_fmin": 0,
+                "mel_fmax": 8000,
+                "freqm": 0,
+                "timem": 0,
+                "blur": False,
+                "mean": -4.63,
+                "std": 2.74,
+                "target_length": 1024,
+            },
+        },
+        "model": {
+            "device": "cuda",
+            "target": "audioldm.pipline.LatentDiffusion",
+            "params": {
+                "base_learning_rate": 5e-06,
+                "linear_start": 0.0015,
+                "linear_end": 0.0195,
+                "num_timesteps_cond": 1,
+                "log_every_t": 200,
+                "timesteps": 1000,
+                "first_stage_key": "fbank",
+                "cond_stage_key": "waveform",
+                "latent_t_size": 256,
+                "latent_f_size": 16,
+                "channels": 8,
+                "cond_stage_trainable": True,
+                "conditioning_key": "film",
+                "monitor": "val/loss_simple_ema",
+                "scale_by_std": True,
+                "unet_config": {
+                    "target": "audioldm.latent_diffusion.openaimodel.UNetModel",
+                    "params": {
+                        "image_size": 64,
+                        "extra_film_condition_dim": 512,
+                        "extra_film_use_concat": True,
+                        "in_channels": 8,
+                        "out_channels": 8,
+                        "model_channels": 128,
+                        "attention_resolutions": [8, 4, 2],
+                        "num_res_blocks": 2,
+                        "channel_mult": [1, 2, 3, 5],
+                        "num_head_channels": 32,
+                        "use_spatial_transformer": True,
+                    },
+                },
+                "first_stage_config": {
+                    "base_learning_rate": 4.5e-05,
+                    "target": "audioldm.variational_autoencoder.autoencoder.AutoencoderKL",
+                    "params": {
+                        "monitor": "val/rec_loss",
+                        "image_key": "fbank",
+                        "subband": 1,
+                        "embed_dim": 8,
+                        "time_shuffle": 1,
+                        "ddconfig": {
+                            "double_z": True,
+                            "z_channels": 8,
+                            "resolution": 256,
+                            "downsample_time": False,
+                            "in_channels": 1,
+                            "out_ch": 1,
+                            "ch": 128,
+                            "ch_mult": [1, 2, 4],
+                            "num_res_blocks": 2,
+                            "attn_resolutions": [],
+                            "dropout": 0.0,
+                        },
+                    },
+                },
+                "cond_stage_config": {
+                    "target": "audioldm.clap.encoders.CLAPAudioEmbeddingClassifierFreev2",
+                    "params": {
+                        "key": "waveform",
+                        "sampling_rate": 16000,
+                        "embed_mode": "audio",
+                        "unconditional_prob": 0.1,
+                    },
+                },
+            },
+        },
+    }
+    
+    if("-l-" in model_name):
+        basic_config["model"]["params"]["unet_config"]["params"]["model_channels"] = 256
+        basic_config["model"]["params"]["unet_config"]["params"]["num_head_channels"] = 64
+    elif("-m-" in model_name):
+        basic_config["model"]["params"]["unet_config"]["params"]["model_channels"] = 192
+        basic_config["model"]["params"]["cond_stage_config"]["params"]["amodel"] = "HTSAT-base" # This model use a larger HTAST
+        
+    return basic_config