mirror of
https://github.com/rakuri255/UltraSinger.git
synced 2026-11-20 13:07:52 +09:00
- Optimized silent recognition
This commit is contained in:
1 parent
b255100f7f
commit
0fcd8a057d
2 files changed
+140
-6
No files matched your search
@@ -0,0 +1,82 @@
|
||||
import os
|
||||
import unittest
|
||||
from src.modules.Audio.silence_processing import remove_silence2
|
||||
import pytest
|
||||
from modules.Speech_Recognition.TranscribedData import TranscribedData
|
||||
|
||||
|
||||
class SilenceProcessingTest(unittest.TestCase):
|
||||
def test_remove_silence2(self):
|
||||
# Arrange
|
||||
timeList = [(2.0, 9.0)]
|
||||
transcribed_data = [
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "Before silence",
|
||||
"end": 2.0,
|
||||
"start": 1.0
|
||||
}),
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "End is in silence",
|
||||
"end": 5.0,
|
||||
"start": 1.0
|
||||
}),
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "Is in silence",
|
||||
"end": 4.0,
|
||||
"start": 3.0
|
||||
}),
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "Start is in silence",
|
||||
"end": 10.0,
|
||||
"start": 5.0
|
||||
}),
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "After silence",
|
||||
"end": 10.0,
|
||||
"start": 9.0
|
||||
}),
|
||||
TranscribedData({
|
||||
"conf": 0.95,
|
||||
"word": "Silence is inside",
|
||||
"end": 12.0,
|
||||
"start": 1.0
|
||||
}),
|
||||
]
|
||||
|
||||
# Act
|
||||
result = remove_silence2(timeList, transcribed_data)
|
||||
|
||||
# Assert
|
||||
self.assertEqual(result[0].word, "Before silence")
|
||||
self.assertEqual(result[0].start, 1.0)
|
||||
self.assertEqual(result[0].end, 2.0)
|
||||
|
||||
self.assertEqual(result[1].word, "End is in silence")
|
||||
self.assertEqual(result[1].start, 1.0)
|
||||
self.assertEqual(result[1].end, 2.0)
|
||||
|
||||
self.assertEqual(len(result), 6) # "in silence" was removed
|
||||
|
||||
self.assertEqual(result[2].word, "Start is in silence")
|
||||
self.assertEqual(result[2].start, 9.0)
|
||||
self.assertEqual(result[2].end, 10.0)
|
||||
|
||||
self.assertEqual(result[3].word, "After silence")
|
||||
self.assertEqual(result[3].start, 9.0)
|
||||
self.assertEqual(result[3].end, 10.0)
|
||||
|
||||
#self.assertEqual(transcribed_data[3].word, "Silence is inside")
|
||||
self.assertEqual(result[4].start, 1.0)
|
||||
self.assertEqual(result[4].end, 2.0)
|
||||
|
||||
#self.assertEqual(transcribed_data[3].word, "Silence is inside")
|
||||
self.assertEqual(result[5].start, 9.0)
|
||||
self.assertEqual(result[5].end, 12.0)
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -1,6 +1,7 @@
|
||||
"""Silence processing module"""
|
||||
|
||||
import librosa
|
||||
from pydub import AudioSegment, silence
|
||||
|
||||
from modules.console_colors import ULTRASINGER_HEAD
|
||||
from modules.Speech_Recognition.TranscribedData import TranscribedData
|
||||
@@ -13,8 +14,61 @@ def remove_silence_from_transcription_data(audio_path: str, transcribed_data: li
|
||||
f"{ULTRASINGER_HEAD} Removing silent start and ending, from transcription data"
|
||||
)
|
||||
|
||||
audio, sample_rate = librosa.load(audio_path, sr=None)
|
||||
#audio, sample_rate = librosa.load(audio_path, sr=None)
|
||||
y = AudioSegment.from_wav(audio_path)
|
||||
silence_timestamps = get_silence_sections(y)
|
||||
a = remove_silence2(silence_timestamps, transcribed_data)
|
||||
#remove_silence(audio, sample_rate, transcribed_data)
|
||||
|
||||
return a
|
||||
|
||||
def get_silence_sections(y,
|
||||
min_silence_len=50,
|
||||
silence_thresh=-50) -> list[tuple[float, float]]:
|
||||
|
||||
s = silence.detect_silence(y, min_silence_len=min_silence_len, silence_thresh=silence_thresh)
|
||||
s = [((start / 1000), (stop / 1000)) for start, stop in s] # convert to sec
|
||||
return s
|
||||
|
||||
def remove_silence2(timeList: list[tuple[float, float]], transcribed_data: list[TranscribedData]):
|
||||
|
||||
new_transcribed_data = []
|
||||
for i, data in enumerate(transcribed_data):
|
||||
new_transcribed_data.append(data)
|
||||
|
||||
for j in range(len(timeList)):
|
||||
silence_start = timeList[j][0]
|
||||
silence_end = timeList[j][1]
|
||||
|
||||
if silence_start >= data.end or silence_end <= data.start:
|
||||
continue
|
||||
if silence_start > data.start and silence_end < data.end:
|
||||
print(f"{ULTRASINGER_HEAD} Splitting \"{data.word}\" because it was in silence")
|
||||
splitted = TranscribedData({
|
||||
"conf": data.conf,
|
||||
"word": "~",
|
||||
"end": data.end,
|
||||
"start": silence_end
|
||||
})
|
||||
data.end = silence_start
|
||||
new_transcribed_data.append(splitted)
|
||||
break
|
||||
if silence_start < data.start and silence_end > data.end:
|
||||
new_transcribed_data.remove(data)
|
||||
print(f"{ULTRASINGER_HEAD} Removed \"{data.word}\" because it was in silence")
|
||||
break
|
||||
if silence_start < data.start:
|
||||
data.start = silence_end
|
||||
print(f"{ULTRASINGER_HEAD} Removed silence in the start of word \"{data.word}\"")
|
||||
if silence_end > data.end:
|
||||
data.end = silence_start
|
||||
print(f"{ULTRASINGER_HEAD} Removed silence in the end of word \"{data.word}\"")
|
||||
if silence_start > data.end:
|
||||
break
|
||||
return new_transcribed_data
|
||||
|
||||
|
||||
def remove_silence(audio, sample_rate, transcribed_data: list[TranscribedData]):
|
||||
for i, data in enumerate(transcribed_data):
|
||||
start_time = data.start
|
||||
end_time = data.end
|
||||
@@ -45,10 +99,8 @@ def remove_silence_from_transcription_data(audio_path: str, transcribed_data: li
|
||||
last_silence_duration = 0
|
||||
|
||||
data.start = (
|
||||
data.start + first_silence_duration
|
||||
data.start + first_silence_duration
|
||||
)
|
||||
data.end = (
|
||||
data.end - last_silence_duration
|
||||
)
|
||||
|
||||
return transcribed_data
|
||||
data.end - last_silence_duration
|
||||
)
|
||||
Reference in new issue
Block a user