finish naive version

.gitignore

@@ -127,3 +127,6 @@ dmypy.json
 
 # Pyre type checker
 .pyre/
+
+# PyCharm
+/.idea/

slicer.py

@@ -0,0 +1,86 @@
+import librosa
+import numpy as np
+import soundfile
+
+
+def _const_conv1d(arr, k_sz, k_val):
+    result = np.zeros(arr.shape[0] - k_sz + 1)
+    total = np.sum(arr[: k_sz])
+    result[0] = total * k_val
+    for i in range(1, result.shape[0]):
+        total = total - arr[i] + arr[i + k_sz - 1]
+        result[i] = total * k_val
+    return result
+    # return np.convolve(arr, np.full(k_sz, k_val), mode='valid')
+
+
+def _rms(audio, win_sz):
+    return np.sqrt(_const_conv1d(np.power(audio, 2), win_sz, 1 / win_sz) - np.power(_const_conv1d(audio, win_sz, 1 / win_sz), 2))
+
+
+def _apply_slice(audio, begin, end):
+    if len(audio.shape) > 1:
+        return audio[:, begin: end]
+    else:
+        return audio[begin: end]
+
+
+class Slicer:
+    def __init__(self,
+                 sr: int,
+                 db_threshold: float = -36,
+                 min_length: int = 5000,
+                 win_l: int = 300,
+                 win_s: int = 20,
+                 max_silence_kept: int = 1000):
+        self.db_threshold = db_threshold
+        self.min_samples = round(sr * min_length / 1000)
+        self.win_ln = round(sr * win_l / 1000)
+        self.win_sn = round(sr * win_s / 1000)
+        if not min_length >= win_l >= win_s:
+            raise ValueError('The following condition must be satisfied: min_length >= win_l >= win_s')
+
+    def slice(self, audio):
+        if len(audio.shape) > 1:
+            samples = librosa.to_mono(audio)
+        else:
+            samples = audio
+        # calculate RMS of each window
+        rms_db_l = np.log10(np.clip(_rms(samples, win_sz=self.win_ln), a_min=1e-12, a_max=1)) * 20
+        rms_db_s = np.log10(np.clip(_rms(samples, win_sz=self.win_sn), a_min=1e-12, a_max=1)) * 20
+        tags = []
+        left = right = 0
+        while right < rms_db_l.shape[0]:
+            if rms_db_l[right] < self.db_threshold:
+                right += 1
+            elif left == right:
+                left += 1
+                right += 1
+            else:
+                # TODO: discard silence parts
+                split_win_m = left + np.argmin(rms_db_s[left: right + self.win_ln - self.win_sn])
+                split_loc_m = split_win_m + np.argmin(rms_db_s[split_win_m: split_win_m + self.win_sn])
+                if 0 < split_loc_m < samples.shape[0] - 1 and (len(tags) == 0 or split_loc_m - tags[-1] >= self.min_samples):
+                    tags.append(split_loc_m)
+                right += 1
+                left = right
+        if len(tags) == 0:
+            return [audio]
+        else:
+            chunks = [_apply_slice(audio, 0, tags[0])]
+            for i in range(1, len(tags)):
+                chunks.append(_apply_slice(audio, tags[i - 1], tags[i]))
+            chunks.append(_apply_slice(audio, tags[-1], samples.shape[0]))
+            return chunks
+
+
+def main():
+    audio, sr = librosa.load(r'D:\Vocoder Datasets\颜绮萱\颜绮萱-2021干声\大鱼-速度140.wav', sr=None)
+    slicer = Slicer(sr=sr, db_threshold=-40, min_length=5000, win_l=750, win_s=20)
+    chunks = slicer.slice(audio)
+    for i, chunk in enumerate(chunks):
+        soundfile.write(fr'D:\Vocoder Datasets\slices\大鱼_{i}.wav', chunk, sr)
+
+
+if __name__ == '__main__':
+    main()