exps

conf/interface-c2f-exp.yml

@@ -0,0 +1,5 @@
+Interface.coarse_ckpt: /runs/c2f-exp-03.22.23/ckpt/mask/epoch=400/vampnet/weights.pth
+Interface.coarse2fine_ckpt: runs/c2f-exp-03.22.23/ckpt/mask/epoch=400/vampnet/weights.pth
+Interface.codec_ckpt: /runs/codec-ckpt/codec.pth
+Interface.coarse_chunk_size_s: 5
+Interface.coarse2fine_chunk_size_s: 3

scripts/exp/{c2f_eval.py → eval.py}

@@ -13,7 +13,7 @@ from audiotools import AudioSignal
 @argbind.bind(without_prefix=True)
 def eval(
     exp_dir: str = None,
-    baseline_key: str = "reconstructed", 
+    baseline_key: str = "baseline", 
     audio_ext: str = ".wav",
 ):
     assert exp_dir is not None
@@ -27,7 +27,7 @@ def eval(
     frechet = FrechetAudioDistance(
         use_pca=False, 
         use_activation=False,
-        verbose=False
+        verbose=True
     )
     visqol = partial(audiotools.metrics.quality.visqol, mode="audio")
 
@@ -48,7 +48,7 @@ def eval(
         cond_dir = exp_dir / condition
         cond_files = list(cond_dir.glob(f"*{audio_ext}"))
 
-        print(f"computing fad")
+        print(f"computing fad for {baseline_dir} and {cond_dir}")
         frechet_score = frechet.score(baseline_dir, cond_dir)
 
         # make sure we have the same number of files

scripts/utils/vamp_folder.py

@@ -5,17 +5,30 @@ from tqdm import tqdm
 import torch
 
 from vampnet.interface import Interface
+import audiotools as at
 
 Interface = argbind.bind(Interface)
 
+# condition wrapper for printing
+def condition(cond):
+    def wrapper(sig, interface):
+        print(f"Condition: {cond.__name__}")
+        sig = cond(sig, interface)
+        print(f"Condition: {cond.__name__} (done)\n")
+        return sig
+    return wrapper
+
+@condition
 def baseline(sig, interface):
     return sig
 
+@condition
 def reconstructed(sig, interface):
     return interface.to_signal(
         interface.encode(sig)
     )
 
+@condition
 def coarse2fine(sig, interface):
     z = interface.encode(sig)
     z = z[:, :interface.c2f.n_conditioning_codebooks, :]
@@ -23,6 +36,18 @@ def coarse2fine(sig, interface):
     z = interface.coarse_to_fine(z)
     return interface.to_signal(z)
 
+@condition
+def coarse2fine_argmax(sig, interface):
+    z = interface.encode(sig)
+    z = z[:, :interface.c2f.n_conditioning_codebooks, :]
+
+    z = interface.coarse_to_fine(z, 
+        sample="argmax", sampling_steps=1, 
+        temperature=1.0
+    )
+    return interface.to_signal(z)
+
+@condition
 def one_codebook(sig, interface):
     z = interface.encode(sig)
 
@@ -38,6 +63,7 @@ def one_codebook(sig, interface):
 
     return interface.to_signal(zv)
 
+@condition
 def four_codebooks_downsampled_4x(sig, interface):
     zv = interface.coarse_vamp_v2(
         sig, downsample_factor=4
@@ -45,6 +71,7 @@ def four_codebooks_downsampled_4x(sig, interface):
     zv = interface.coarse_to_fine(zv)  
     return interface.to_signal(zv)
 
+@condition
 def two_codebooks_downsampled_4x(sig, interface):
     z = interface.encode(sig)
 
@@ -60,6 +87,7 @@ def two_codebooks_downsampled_4x(sig, interface):
 
     return interface.to_signal(zv)
 
+@condition
 def four_codebooks_downsampled_8x(sig, interface):
     zv = interface.coarse_vamp_v2(
         sig, downsample_factor=8
@@ -68,7 +96,7 @@ def four_codebooks_downsampled_8x(sig, interface):
     return interface.to_signal(zv)
 
 
-SAMPLE_CONDS ={
+COARSE_SAMPLE_CONDS ={
     "baseline": baseline,
     "reconstructed": reconstructed,
     "coarse2fine": coarse2fine,
@@ -78,6 +106,12 @@ SAMPLE_CONDS ={
     "four_codebooks_downsampled_8x": four_codebooks_downsampled_8x,
 }
 
+C2F_SAMPLE_CONDS = {
+    "baseline": baseline,
+    "reconstructed": reconstructed,
+    "coarse2fine": coarse2fine,
+    "coarse2fine_argmax": coarse2fine_argmax,
+}
 
 @argbind.bind(without_prefix=True)
 def main(
@@ -86,7 +120,10 @@ def main(
         ], 
         output_dir: str = "./samples",
         max_excerpts: int = 5000,
+        exp_type: str = "coarse", 
+        seed: int = 0,
     ):
+    at.util.seed(seed)
     interface = Interface()
 
     output_dir = Path(output_dir) 
@@ -102,6 +139,8 @@ def main(
         without_replacement=True,
     )
 
+    SAMPLE_CONDS = COARSE_SAMPLE_CONDS if exp_type == "coarse" else C2F_SAMPLE_CONDS
+
     for i in tqdm(range(max_excerpts)):
         sig = dataset[i]["signal"]