add player script and fix bug

data.py

@@ -6,7 +6,15 @@ import mlflow
 
 SAMPLE_RATE = 22050
 
-#@mlflow.trace
+def spec_to_audio(spec):
+    """
+    Convert a normalized mel-spectrogram back to audio.
+    """
+    spec = (spec * 80) - 80
+    spec = librosa.db_to_amplitude(spec)*80
+    audio = librosa.feature.inverse.mel_to_audio(spec,sr=SAMPLE_RATE)
+    return audio
+
 def process_file(file_path):
     """
     Load 10 second chunks single song.

run.py

@@ -0,0 +1,43 @@
+import numpy as np
+import random
+import time
+from tinygrad import Tensor, nn
+from tinygrad.nn.state import safe_load, load_state_dict
+import librosa
+import sounddevice as sd
+from model import gen
+from data import spec_to_audio
+
+SAMPLE_RATE = 22050
+
+def load_model(filepath="model.safetensors"):
+    """Loads the model structure and weights."""
+    model = gen()
+    state_dict = safe_load(filepath)
+    load_state_dict(model, state_dict)
+    return model
+
+def load_data(filepath="data.npz"):
+    """Loads the pre-processed spectrogram data."""
+    print(f"Loading data from {filepath}...")
+    data = np.load(filepath)
+    x = data["arr_0"]
+    return x
+
+def play_spec(spec,i):
+    """Converts a spectrogram numpy array to audio and plays it."""
+    audio = spec_to_audio(spec)
+    sd.wait()
+    print(f"chunk:{i}")
+    sd.play(audio, samplerate=SAMPLE_RATE)
+
+def run_prediction_loop(model, data_x):
+    current_spect = data_x[0:1]
+    for i in range(10):
+        play_spec(current_spect[0][0],i)
+        current_spect = model(Tensor(current_spect)).numpy()
+
+if __name__ == "__main__":
+    model = load_model()
+    data_x = load_data()
+    run_prediction_loop(model, data_x)

train.py

@@ -1,6 +1,7 @@
 import mlflow
 import numpy as np
 from tinygrad import Device,Tensor,nn,TinyJit
+from tinygrad.nn.state import safe_save, get_state_dict
 import matplotlib.pyplot as plt
 import time
 import show
@@ -43,8 +44,7 @@ mlflow.log_params({"batch_size": BATCH_SIZE, "epochs": EPOCHS, "lr": LEARNING_RA
 show.logSpec(Tensor(x[0:1]).numpy()[0][0],"default")
 
 print("training")
-pl = 0
+eshape = (BATCH_SIZE, 1, 128, 216)
-eshape = (BATCH_SIZE, 1, 128, 431)
 for epoch in range(0,EPOCHS):
     print(f"\n--- Starting Epoch {epoch} ---\n")
     loss=0
@@ -65,3 +65,7 @@ for epoch in range(0,EPOCHS):
 
     mlflow.log_metric("loss", loss, step=epoch)
     print(f"loss of {loss}")
+
+show.logSpec(mdl(Tensor(x[0:1])).numpy()[0][0],EPOCHS)
+state_dict = get_state_dict(mdl)
+safe_save(state_dict, "model.safetensors")