cuda

predict.py

@@ -129,11 +129,15 @@ def predict(
             default="peak",
             choices=["loudness", "clip", "peak", "rms"],
         ),
+        beat_sync_threshold: float = Input(
+            description="When beat syncing, if the gap between generated downbeat timing and input audio downbeat timing is larger than `beat_sync_threshold`, consider the beats are not corresponding. (This will be fixed with the optimal value and be hidden, when releasing.)",
+            default=0.75,
+        ),
         chroma_coefficient: float = Input(
             description="Coefficient value multiplied to multi-hot chord chroma.",
             default=1.0,
             ge=0.5,
-            le=2.5
+            le=2.0
         ),
         top_k: int = Input(
             description="Reduces sampling to the k most likely tokens.", default=250
@@ -159,22 +163,18 @@ def predict(
             description="Seed for random number generator. If `None` or `-1`, a random seed will be used.",
             default=None,
         ),
-        overlap: int = Input(
-            description="The length of overlapping part. Last `overlap` seconds of previous generation output audio is given to the next generation's audio prompt for continuation. (This will be fixed with the optimal value and be hidden, when releasing.)",
-            default=5, le=15, ge=1
-        ),
-        in_step_beat_sync: bool = Input(
-            description="If `True`, beat syncing is performed every generation step. In this case, audio prompting with EnCodec token will not be used, so that the audio quality might be degraded on and on along encoding-decoding sequences of the generation steps. (This will be fixed with the optimal value and be hidden, when releasing.)",
-            default=False,
-        ),
-        beat_sync_threshold: float = Input(
-            description="When beat syncing, if the gap between generated downbeat timing and input audio downbeat timing is larger than `beat_sync_threshold`, consider the beats are not corresponding. (This will be fixed with the optimal value and be hidden, when releasing.)",
-            default=0.75,
-        ),
-        amp_rate: float = Input(
-            description="Amplifying the output audio to prevent volume diminishing along generations. (This will be fixed with the optimal value and be hidden, when releasing.)",
-            default=1.2,
-        ),
+        # overlap: int = Input(
+        #     description="The length of overlapping part. Last `overlap` seconds of previous generation output audio is given to the next generation's audio prompt for continuation. (This will be fixed with the optimal value and be hidden, when releasing.)",
+        #     default=5, le=15, ge=1
+        # ),
+        # in_step_beat_sync: bool = Input(
+        #     description="If `True`, beat syncing is performed every generation step. In this case, audio prompting with EnCodec token will not be used, so that the audio quality might be degraded on and on along encoding-decoding sequences of the generation steps. (This will be fixed with the optimal value and be hidden, when releasing.)",
+        #     default=False,
+        # ),
+        # amp_rate: float = Input(
+        #     description="Amplifying the output audio to prevent volume diminishing along generations. (This will be fixed with the optimal value and be hidden, when releasing.)",
+        #     default=1.2,
+        # ),
     ) -> Path:
 
         if prompt is None:
@@ -201,7 +201,7 @@ def predict(
         model = self.model
         model.lm.eval()
 
-        in_step_beat_sync = in_step_beat_sync
+        # in_step_beat_sync = in_step_beat_sync
 
         set_generation_params = lambda duration: model.set_generation_params(
             duration=duration,
@@ -241,7 +241,7 @@ def predict(
         )
 
         beat_sync_threshold = beat_sync_threshold
-        amp_rate = amp_rate
+        # amp_rate = amp_rate
 
         set_generation_params(duration)
 
@@ -277,7 +277,7 @@ def predict(
 
         downbeat_offset = input_downbeats[0]-wav_downbeats[0]
         if downbeat_offset > 0:
-            wav = torch.concat([torch.zeros([1,1,int(downbeat_offset*wav_sr)]),wav],dim=-1)
+            wav = torch.concat([torch.zeros([1,1,int(downbeat_offset*wav_sr)]).cpu(),wav.cpu()],dim=-1)
             for i in range(len(wav_downbeats)):
                 wav_downbeats[i]=wav_downbeats[i]+downbeat_offset
         wav_downbeats = [0] + wav_downbeats + [wav_length]