Optimize silence detection

This commit is contained in:
Vadim committed 2023-12-27 22:16:05 +01:00
1 parent bff2d336e2
commit dad84ebdea
4 files changed
+60 -18

No files matched your search

+26 -11
View File
@@ -66,31 +66,37 @@ class SilenceProcessingTest(unittest.TestCase):
self.assertEqual(result[0].word, "Before silence ")
self.assertEqual(result[0].start, 1.0)
self.assertEqual(result[0].end, 2.0)
self.assertEqual(result[0].is_hyphen, None)
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(result[1].is_hyphen, None)
self.assertEqual(len(result), 6) # "in silence" is 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[2].is_hyphen, None)
self.assertEqual(result[3].word, "After silence ")
self.assertEqual(result[3].start, 9.0)
self.assertEqual(result[3].end, 10.0)
self.assertEqual(result[3].is_hyphen, None)
# Split to "Silence is inside~"
# Split to "Silence is inside~ "
self.assertEqual(result[4].word, "Silence is inside")
self.assertEqual(result[4].start, 1.0)
self.assertEqual(result[4].end, 2.0)
self.assertEqual(result[4].is_hyphen, True)
self.assertEqual(result[5].word, "~ ")
self.assertEqual(result[5].start, 9.0)
self.assertEqual(result[5].end, 12.0)
self.assertEqual(result[5].is_hyphen, False)
def test_remove_multi_silence2(self):
def test_remove_multiple_silence_in_between_silence2(self):
#
# | ** ** ** ** | silence_parts_list
# | 00**00**00**00**00000000 | multi silence is inside
@@ -115,25 +121,30 @@ class SilenceProcessingTest(unittest.TestCase):
self.assertEqual(result[0].word, "Silence is inside")
self.assertEqual(result[0].start, 1.0)
self.assertEqual(result[0].end, 2.0)
self.assertEqual(result[0].is_hyphen, True)
self.assertEqual(result[1].word, "1")
self.assertEqual(result[1].word, "~")
self.assertEqual(result[1].start, 3.0)
self.assertEqual(result[1].end, 4.0)
self.assertEqual(result[1].is_hyphen, True)
self.assertEqual(result[2].word, "1")
self.assertEqual(result[2].word, "~")
self.assertEqual(result[2].start, 5.0)
self.assertEqual(result[2].end, 6.0)
self.assertEqual(result[2].is_hyphen, True)
self.assertEqual(result[3].word, "1")
self.assertEqual(result[3].word, "~")
self.assertEqual(result[3].start, 7.0)
self.assertEqual(result[3].end, 8.0)
self.assertEqual(result[3].is_hyphen, True)
self.assertEqual(result[4].word, "2 ")
self.assertEqual(result[4].word, "~ ")
self.assertEqual(result[4].start, 9.0)
self.assertEqual(result[4].end, 12.0)
self.assertEqual(result[4].is_hyphen, False)
def test_remove_multi_silence2(self):
def test_remove_multiple_silence_till_end_silence2(self):
#
# | ** ** ** ** | silence_parts_list
# | 00**00**00**00** | multi silence is inside
@@ -158,22 +169,26 @@ class SilenceProcessingTest(unittest.TestCase):
self.assertEqual(result[0].word, "Silence is inside")
self.assertEqual(result[0].start, 1.0)
self.assertEqual(result[0].end, 2.0)
self.assertEqual(result[0].is_hyphen, True)
self.assertEqual(result[1].word, "1")
self.assertEqual(result[1].word, "~")
self.assertEqual(result[1].start, 3.0)
self.assertEqual(result[1].end, 4.0)
self.assertEqual(result[1].is_hyphen, True)
self.assertEqual(result[2].word, "1")
self.assertEqual(result[2].word, "~")
self.assertEqual(result[2].start, 5.0)
self.assertEqual(result[2].end, 6.0)
self.assertEqual(result[2].is_hyphen, True)
self.assertEqual(result[3].word, "2 ")
self.assertEqual(result[3].word, "~ ")
self.assertEqual(result[3].start, 7.0)
self.assertEqual(result[3].end, 8.0)
self.assertEqual(result[3].is_hyphen, False)
# todo
# | ** ** ** ***** | silence_parts_list
# | 00**00**00**0 | multi silence is inside
# | *00**00**0 | multi silence is inside
# |0 1 2 3 4 5 6 7 8 9 10 11 12| time
if __name__ == "__main__":
unittest.main()
+9 -5
View File
@@ -58,18 +58,22 @@ def remove_silence2(silence_parts_list: list[tuple[float, float]], transcribed_d
if next_index < len(silence_parts_list) and silence_parts_list[next_index][0] < origin_end:
split_end = silence_parts_list[next_index][0]
if silence_parts_list[next_index][1] > origin_end:
split_word = "2 "
if silence_parts_list[next_index][1] >= origin_end:
split_word = "~ "
is_hyphen = False
else:
split_word = "1"
split_word = "~"
is_hyphen = True
else:
split_end = origin_end
split_word = "2 "
split_word = "~ "
is_hyphen = False
split_data = TranscribedData({"conf": data.conf, "word": split_word, "end": split_end, "start": silence_end})
split_data = TranscribedData({"conf": data.conf, "word": split_word, "end": split_end, "start": silence_end, "is_hyphen": is_hyphen})
if not was_split:
data.end = silence_start
data.is_hyphen = True
# Remove last whitespace from the data.word
if data.word[-1] == " ":
data.word = data.word[:-1]
@@ -15,4 +15,4 @@ class TranscribedData:
)
self.end = transcribed_json.get("end", None)
self.start = transcribed_json.get("start", None)
self.is_hyphen = None
self.is_hyphen = transcribed_json.get("is_hyphen", None)
+24 -1
View File
@@ -14,6 +14,25 @@ from modules.Ultrastar.ultrastar_txt import UltrastarTxtValue, UltrastarTxtTag,
from modules.Speech_Recognition.TranscribedData import TranscribedData
from modules.Ultrastar.ultrastar_score_calculator import Score
def get_sixteenth_note_second(real_bpm: float):
"""Converts a beat to a 1/16 note in second"""
return 60 / real_bpm / 4
def get_eighth_note_second(real_bpm: float):
"""Converts a beat to a 1/8 note in second"""
return 60 / real_bpm / 2
def get_quarter_note_second(real_bpm: float):
"""Converts a beat to a 1/4 note in second"""
return 60 / real_bpm
def get_half_note_second(real_bpm: float):
"""Converts a beat to a 1/2 note in second"""
return 60 / real_bpm * 2
def get_whole_note_second(real_bpm: float):
"""Converts a beat to a 1/1 note in second"""
return 60 / real_bpm * 4
def get_multiplier(real_bpm: float) -> int:
"""Calculates the multiplier for the BPM"""
@@ -51,6 +70,7 @@ def create_ultrastar_txt_from_automation(
ultrastar_bpm = real_bpm_to_ultrastar_bpm(bpm)
multiplication = get_multiplier(ultrastar_bpm)
ultrastar_bpm = ultrastar_bpm * get_multiplier(ultrastar_bpm)
silence_split_duration = get_eighth_note_second(bpm)
with open(ultrastar_file_output, "w", encoding=FILE_ENCODING) as file:
gap = transcribed_data[0].start
@@ -101,6 +121,9 @@ def create_ultrastar_txt_from_automation(
file.write(line)
# detect silence between words
if transcribed_data[i].is_hyphen:
continue
if i < len(transcribed_data) - 1:
silence = (
transcribed_data[i + 1].start - data.end
@@ -108,7 +131,7 @@ def create_ultrastar_txt_from_automation(
else:
silence = 0
if silence > 0.3 and i != len(transcribed_data) - 1:
if silence > silence_split_duration and i != len(transcribed_data) - 1:
# - 10
# '-' end of current sing part
# 'n1' show next at time in real beat