merge try

app.py

@@ -251,102 +251,87 @@ def circular_padding(spec: np.ndarray, end: int):
 
 
 def wav2mel(audio_path: str, width=2, top_db=40):
-    os.makedirs(TEMP_DIR, exist_ok=True)
-    try:
-        y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
-        non_silents = librosa.effects.split(y, top_db=top_db)
-        y = np.concatenate([y[start:end] for start, end in non_silents])
-        total_frames = len(y)
-        if total_frames % (width * sr) != 0:
-            count = total_frames // (width * sr) + 1
-            y = circular_padding(y, count * width * sr)
-
-        mel_spec = librosa.feature.melspectrogram(y=y, sr=sr)
-        log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max)
-        dur = librosa.get_duration(y=y, sr=sr)
-        total_frames = log_mel_spec.shape[1]
-        step = int(width * total_frames / dur)
-        count = int(total_frames / step)
-        begin = int(0.5 * (total_frames - count * step))
-        end = begin + step * count
-        for i in range(begin, end, step):
-            librosa.display.specshow(log_mel_spec[:, i : i + step])
-            plt.axis("off")
-            plt.savefig(
-                f"{TEMP_DIR}/{i}.jpg",
-                bbox_inches="tight",
-                pad_inches=0.0,
-            )
-            plt.close()
+    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
+    non_silents = librosa.effects.split(y, top_db=top_db)
+    y = np.concatenate([y[start:end] for start, end in non_silents])
+    total_frames = len(y)
+    if total_frames % (width * sr) != 0:
+        count = total_frames // (width * sr) + 1
+        y = circular_padding(y, count * width * sr)
 
-    except Exception as e:
-        print(f"Error converting {audio_path} : {e}")
+    mel_spec = librosa.feature.melspectrogram(y=y, sr=sr)
+    log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max)
+    dur = librosa.get_duration(y=y, sr=sr)
+    total_frames = log_mel_spec.shape[1]
+    step = int(width * total_frames / dur)
+    count = int(total_frames / step)
+    begin = int(0.5 * (total_frames - count * step))
+    end = begin + step * count
+    for i in range(begin, end, step):
+        librosa.display.specshow(log_mel_spec[:, i : i + step])
+        plt.axis("off")
+        plt.savefig(
+            f"{TEMP_DIR}/{i}.jpg",
+            bbox_inches="tight",
+            pad_inches=0.0,
+        )
+        plt.close()
 
 
 def wav2cqt(audio_path: str, width=2, top_db=40):
-    os.makedirs(TEMP_DIR, exist_ok=True)
-    try:
-        y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
-        non_silents = librosa.effects.split(y, top_db=top_db)
-        y = np.concatenate([y[start:end] for start, end in non_silents])
-        total_frames = len(y)
-        if total_frames % (width * sr) != 0:
-            count = total_frames // (width * sr) + 1
-            y = circular_padding(y, count * width * sr)
-
-        cqt_spec = librosa.cqt(y=y, sr=sr)
-        log_cqt_spec = librosa.power_to_db(np.abs(cqt_spec) ** 2, ref=np.max)
-        dur = librosa.get_duration(y=y, sr=sr)
-        total_frames = log_cqt_spec.shape[1]
-        step = int(width * total_frames / dur)
-        count = int(total_frames / step)
-        begin = int(0.5 * (total_frames - count * step))
-        end = begin + step * count
-        for i in range(begin, end, step):
-            librosa.display.specshow(log_cqt_spec[:, i : i + step])
-            plt.axis("off")
-            plt.savefig(
-                f"{TEMP_DIR}/{i}.jpg",
-                bbox_inches="tight",
-                pad_inches=0.0,
-            )
-            plt.close()
+    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
+    non_silents = librosa.effects.split(y, top_db=top_db)
+    y = np.concatenate([y[start:end] for start, end in non_silents])
+    total_frames = len(y)
+    if total_frames % (width * sr) != 0:
+        count = total_frames // (width * sr) + 1
+        y = circular_padding(y, count * width * sr)
 
-    except Exception as e:
-        print(f"Error converting {audio_path} : {e}")
+    cqt_spec = librosa.cqt(y=y, sr=sr)
+    log_cqt_spec = librosa.power_to_db(np.abs(cqt_spec) ** 2, ref=np.max)
+    dur = librosa.get_duration(y=y, sr=sr)
+    total_frames = log_cqt_spec.shape[1]
+    step = int(width * total_frames / dur)
+    count = int(total_frames / step)
+    begin = int(0.5 * (total_frames - count * step))
+    end = begin + step * count
+    for i in range(begin, end, step):
+        librosa.display.specshow(log_cqt_spec[:, i : i + step])
+        plt.axis("off")
+        plt.savefig(
+            f"{TEMP_DIR}/{i}.jpg",
+            bbox_inches="tight",
+            pad_inches=0.0,
+        )
+        plt.close()
 
 
 def wav2chroma(audio_path: str, width=2, top_db=40):
-    os.makedirs(TEMP_DIR, exist_ok=True)
-    try:
-        y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
-        non_silents = librosa.effects.split(y, top_db=top_db)
-        y = np.concatenate([y[start:end] for start, end in non_silents])
-        total_frames = len(y)
-        if total_frames % (width * sr) != 0:
-            count = total_frames // (width * sr) + 1
-            y = circular_padding(y, count * width * sr)
-
-        chroma_spec = librosa.feature.chroma_stft(y=y, sr=sr)
-        log_chroma_spec = librosa.power_to_db(np.abs(chroma_spec) ** 2, ref=np.max)
-        dur = librosa.get_duration(y=y, sr=sr)
-        total_frames = log_chroma_spec.shape[1]
-        step = int(width * total_frames / dur)
-        count = int(total_frames / step)
-        begin = int(0.5 * (total_frames - count * step))
-        end = begin + step * count
-        for i in range(begin, end, step):
-            librosa.display.specshow(log_chroma_spec[:, i : i + step])
-            plt.axis("off")
-            plt.savefig(
-                f"{TEMP_DIR}/{i}.jpg",
-                bbox_inches="tight",
-                pad_inches=0.0,
-            )
-            plt.close()
+    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
+    non_silents = librosa.effects.split(y, top_db=top_db)
+    y = np.concatenate([y[start:end] for start, end in non_silents])
+    total_frames = len(y)
+    if total_frames % (width * sr) != 0:
+        count = total_frames // (width * sr) + 1
+        y = circular_padding(y, count * width * sr)
 
-    except Exception as e:
-        print(f"Error converting {audio_path} : {e}")
+    chroma_spec = librosa.feature.chroma_stft(y=y, sr=sr)
+    log_chroma_spec = librosa.power_to_db(np.abs(chroma_spec) ** 2, ref=np.max)
+    dur = librosa.get_duration(y=y, sr=sr)
+    total_frames = log_chroma_spec.shape[1]
+    step = int(width * total_frames / dur)
+    count = int(total_frames / step)
+    begin = int(0.5 * (total_frames - count * step))
+    end = begin + step * count
+    for i in range(begin, end, step):
+        librosa.display.specshow(log_chroma_spec[:, i : i + step])
+        plt.axis("off")
+        plt.savefig(
+            f"{TEMP_DIR}/{i}.jpg",
+            bbox_inches="tight",
+            pad_inches=0.0,
+        )
+        plt.close()
 
 
 def most_frequent_value(lst: list):
@@ -366,13 +351,15 @@ def infer(wav_path: str, log_name: str, folder_path=TEMP_DIR):
     if not wav_path:
         return None, "Please input an audio!"
 
+    spec = log_name.split("_")[-3]
+    os.makedirs(folder_path, exist_ok=True)
     try:
         model = EvalNet(log_name, len(TRANSLATE)).model
+        eval("wav2%s" % spec)(wav_path)
+
     except Exception as e:
         return None, f"{e}"
 
-    spec = log_name.split("_")[-3]
-    eval("wav2%s" % spec)(wav_path)
     jpgs = find_files(folder_path, ".jpg")
     preds = []
     for jpg in jpgs: