Spaces:
Sleeping
Sleeping
Commit
路
96b9863
1
Parent(s):
51fae2f
commmit
Browse files
app.py
CHANGED
|
@@ -1,41 +1,75 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
-
import tensorflow as tf
|
| 3 |
-
import librosa
|
| 4 |
import numpy as np
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
|
| 6 |
# Wczytanie modelu
|
| 7 |
model = tf.keras.models.load_model("model.h5")
|
| 8 |
|
| 9 |
-
|
| 10 |
-
|
|
|
|
|
|
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
y = np.pad(y, (0, SR * DURATION - len(y)))
|
| 21 |
-
mel = librosa.feature.melspectrogram(y=y, sr=SR, n_mels=N_MELS)
|
| 22 |
-
mel_db = librosa.power_to_db(mel, ref=np.max)
|
| 23 |
-
mel_db = mel_db[..., np.newaxis] # Dodaj kana艂
|
| 24 |
-
mel_db = np.expand_dims(mel_db, axis=0) # Dodaj batch
|
| 25 |
-
return mel_db
|
| 26 |
-
|
| 27 |
-
def predict_instrument(audio_file):
|
| 28 |
-
mel_input = preprocess_audio(audio_file)
|
| 29 |
-
preds = model.predict(mel_input)[0]
|
| 30 |
-
result = {cls: float(score) for cls, score in zip(INSTRUMENTS, preds)}
|
| 31 |
-
return result
|
| 32 |
|
|
|
|
| 33 |
demo = gr.Interface(
|
| 34 |
-
fn=
|
| 35 |
inputs=gr.Audio(type="filepath", label="Wgraj plik WAV"),
|
| 36 |
-
outputs=
|
| 37 |
-
|
| 38 |
-
|
|
|
|
|
|
|
|
|
|
| 39 |
)
|
| 40 |
|
| 41 |
if __name__ == "__main__":
|
|
|
|
| 1 |
import gradio as gr
|
|
|
|
|
|
|
| 2 |
import numpy as np
|
| 3 |
+
import librosa
|
| 4 |
+
import librosa.display
|
| 5 |
+
import tensorflow as tf
|
| 6 |
+
import matplotlib.pyplot as plt
|
| 7 |
+
|
| 8 |
+
# Parametry modelu
|
| 9 |
+
SR = 22050
|
| 10 |
+
N_MELS = 128
|
| 11 |
+
TARGET_FRAMES = 216
|
| 12 |
+
LABELS = ['cel', 'cla', 'flu', 'gac', 'gel', 'org', 'pia', 'sax', 'tru', 'vio', 'voi']
|
| 13 |
|
| 14 |
# Wczytanie modelu
|
| 15 |
model = tf.keras.models.load_model("model.h5")
|
| 16 |
|
| 17 |
+
def compute_melspectrogram(y, sr=SR, n_mels=N_MELS):
|
| 18 |
+
S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=n_mels)
|
| 19 |
+
S_DB = librosa.power_to_db(S, ref=np.max)
|
| 20 |
+
return S_DB
|
| 21 |
|
| 22 |
+
def resize_spectrogram(S, target_frames=TARGET_FRAMES):
|
| 23 |
+
if S.shape[1] < target_frames:
|
| 24 |
+
pad = target_frames - S.shape[1]
|
| 25 |
+
left = pad // 2; right = pad - left
|
| 26 |
+
S = np.pad(S, ((0, 0), (left, right)), mode='constant')
|
| 27 |
+
elif S.shape[1] > target_frames:
|
| 28 |
+
start = (S.shape[1] - target_frames) // 2
|
| 29 |
+
S = S[:, start:start+target_frames]
|
| 30 |
+
return S
|
| 31 |
+
|
| 32 |
+
def predict_and_plot(audio_path):
|
| 33 |
+
# Wczytaj audio
|
| 34 |
+
y, _ = librosa.load(audio_path, sr=SR)
|
| 35 |
+
|
| 36 |
+
# Oblicz spektrogram
|
| 37 |
+
S_full = compute_melspectrogram(y)
|
| 38 |
+
S = resize_spectrogram(S_full)
|
| 39 |
+
|
| 40 |
+
# Przygotuj do predykcji
|
| 41 |
+
x = S[np.newaxis, ..., np.newaxis]
|
| 42 |
+
probs = model.predict(x, verbose=0)[0]
|
| 43 |
+
|
| 44 |
+
# Przygotuj spektrogram jako obrazek
|
| 45 |
+
fig, ax = plt.subplots(figsize=(8, 4))
|
| 46 |
+
librosa.display.specshow(S_full, sr=SR, x_axis='time', y_axis='mel', cmap='magma', ax=ax)
|
| 47 |
+
ax.set_title("Mel-spektrogram")
|
| 48 |
+
ax.set_xlabel("Czas")
|
| 49 |
+
ax.set_ylabel("Cz臋stotliwo艣膰 (Mel)")
|
| 50 |
+
plt.tight_layout()
|
| 51 |
+
|
| 52 |
+
# Zapisz obrazek do zmiennej
|
| 53 |
+
import io
|
| 54 |
+
buf = io.BytesIO()
|
| 55 |
+
plt.savefig(buf, format='png')
|
| 56 |
+
plt.close(fig)
|
| 57 |
+
buf.seek(0)
|
| 58 |
|
| 59 |
+
# Zwr贸膰 predykcje i obraz
|
| 60 |
+
predictions = {label: float(p) for label, p in zip(LABELS, probs)}
|
| 61 |
+
return predictions, buf
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
|
| 63 |
+
# Gradio UI
|
| 64 |
demo = gr.Interface(
|
| 65 |
+
fn=predict_and_plot,
|
| 66 |
inputs=gr.Audio(type="filepath", label="Wgraj plik WAV"),
|
| 67 |
+
outputs=[
|
| 68 |
+
gr.Label(num_top_classes=5, label="Predykcje instrumentu"),
|
| 69 |
+
gr.Image(label="Spektrogram")
|
| 70 |
+
],
|
| 71 |
+
title="Rozpoznawanie instrument贸w z d藕wi臋ku",
|
| 72 |
+
description="Model na podstawie spektrogramu melowego rozpoznaje instrument muzyczny. Obs艂ugiwane klasy: " + ", ".join(LABELS)
|
| 73 |
)
|
| 74 |
|
| 75 |
if __name__ == "__main__":
|