Update app.py
Browse files
app.py
CHANGED
@@ -1,5 +1,4 @@
|
|
1 |
import torch
|
2 |
-
import spaces
|
3 |
import gradio as gr
|
4 |
import os
|
5 |
from pyannote.audio import Pipeline
|
@@ -60,7 +59,7 @@ def combine_audio_with_time(target_audio, mixed_audio):
|
|
60 |
return {"start_time": target_start_time, "end_time": target_end_time}
|
61 |
|
62 |
# 使用 pyannote/speaker-diarization 对拼接后的音频进行说话人分离
|
63 |
-
@
|
64 |
def diarize_audio(temp_file):
|
65 |
if pipeline is None:
|
66 |
return "错误: 模型未初始化"
|
@@ -74,7 +73,7 @@ def diarize_audio(temp_file):
|
|
74 |
except Exception as e:
|
75 |
return f"处理音频时出错: {e}"
|
76 |
|
77 |
-
#
|
78 |
def get_speaker_segments(diarization, target_start_time, target_end_time, final_audio_length):
|
79 |
speaker_segments = {}
|
80 |
|
@@ -85,20 +84,17 @@ def get_speaker_segments(diarization, target_start_time, target_end_time, final_
|
|
85 |
|
86 |
# 如果是目标说话人
|
87 |
if speaker == 'SPEAKER_00':
|
88 |
-
#
|
89 |
if start < target_end_time and end > target_start_time:
|
90 |
-
#
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
# 完全不与目标音频重叠的时间段
|
100 |
-
if end <= target_start_time or start >= target_end_time:
|
101 |
-
speaker_segments.setdefault(speaker, []).append((start, end))
|
102 |
|
103 |
return speaker_segments
|
104 |
|
@@ -122,7 +118,7 @@ def process_audio(target_audio, mixed_audio):
|
|
122 |
# 获取拼接后的音频长度
|
123 |
final_audio_length = len(AudioSegment.from_wav("final_output.wav")) / 1000 # 秒为单位
|
124 |
|
125 |
-
#
|
126 |
speaker_segments = get_speaker_segments(
|
127 |
diarization_result,
|
128 |
time_dict['start_time'],
|
@@ -162,4 +158,4 @@ with gr.Blocks() as demo:
|
|
162 |
outputs=[diarization_output]
|
163 |
)
|
164 |
|
165 |
-
demo.launch(share=True)
|
|
|
1 |
import torch
|
|
|
2 |
import gradio as gr
|
3 |
import os
|
4 |
from pyannote.audio import Pipeline
|
|
|
59 |
return {"start_time": target_start_time, "end_time": target_end_time}
|
60 |
|
61 |
# 使用 pyannote/speaker-diarization 对拼接后的音频进行说话人分离
|
62 |
+
@gr.Interface(duration=60 * 2) # 使用 GPU 加速,限制执行时间为 120 秒
|
63 |
def diarize_audio(temp_file):
|
64 |
if pipeline is None:
|
65 |
return "错误: 模型未初始化"
|
|
|
73 |
except Exception as e:
|
74 |
return f"处理音频时出错: {e}"
|
75 |
|
76 |
+
# 获取目标说话人的时间段并替换指定的SPEAKER_00
|
77 |
def get_speaker_segments(diarization, target_start_time, target_end_time, final_audio_length):
|
78 |
speaker_segments = {}
|
79 |
|
|
|
84 |
|
85 |
# 如果是目标说话人
|
86 |
if speaker == 'SPEAKER_00':
|
87 |
+
# 替换目标音频的时间段
|
88 |
if start < target_end_time and end > target_start_time:
|
89 |
+
# 目标音频时间段被截断,重新计算其时间段
|
90 |
+
new_start = max(start, target_start_time)
|
91 |
+
new_end = min(end, target_end_time)
|
92 |
+
speaker_segments.setdefault(speaker, []).append((new_start, new_end))
|
93 |
+
|
94 |
+
else:
|
95 |
+
# 完全不与目标音频重叠的时间段
|
96 |
+
if end <= target_start_time or start >= target_end_time:
|
97 |
+
speaker_segments.setdefault(speaker, []).append((start, end))
|
|
|
|
|
|
|
98 |
|
99 |
return speaker_segments
|
100 |
|
|
|
118 |
# 获取拼接后的音频长度
|
119 |
final_audio_length = len(AudioSegment.from_wav("final_output.wav")) / 1000 # 秒为单位
|
120 |
|
121 |
+
# 获取目标说话人的时间段(已排除和截断目标音频时间段)
|
122 |
speaker_segments = get_speaker_segments(
|
123 |
diarization_result,
|
124 |
time_dict['start_time'],
|
|
|
158 |
outputs=[diarization_output]
|
159 |
)
|
160 |
|
161 |
+
demo.launch(share=True)
|