Spaces:
Running
Running
jhj0517
commited on
Commit
·
ddbe0b6
1
Parent(s):
78d8e18
Apply Segment model to the pipeline
Browse files- modules/diarize/diarize_pipeline.py +1 -0
- modules/diarize/diarizer.py +16 -8
- modules/utils/subtitle_manager.py +11 -0
- modules/vad/silero_vad.py +6 -5
- modules/whisper/faster_whisper_inference.py +8 -8
- modules/whisper/insanely_fast_whisper_inference.py +14 -6
- modules/whisper/whisper_Inference.py +24 -17
modules/diarize/diarize_pipeline.py
CHANGED
@@ -44,6 +44,7 @@ class DiarizationPipeline:
|
|
44 |
def assign_word_speakers(diarize_df, transcript_result, fill_nearest=False):
|
45 |
transcript_segments = transcript_result["segments"]
|
46 |
for seg in transcript_segments:
|
|
|
47 |
# assign speaker to segment (if any)
|
48 |
diarize_df['intersection'] = np.minimum(diarize_df['end'], seg['end']) - np.maximum(diarize_df['start'],
|
49 |
seg['start'])
|
|
|
44 |
def assign_word_speakers(diarize_df, transcript_result, fill_nearest=False):
|
45 |
transcript_segments = transcript_result["segments"]
|
46 |
for seg in transcript_segments:
|
47 |
+
seg = seg.dict()
|
48 |
# assign speaker to segment (if any)
|
49 |
diarize_df['intersection'] = np.minimum(diarize_df['end'], seg['end']) - np.maximum(diarize_df['start'],
|
50 |
seg['start'])
|
modules/diarize/diarizer.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import os
|
2 |
import torch
|
3 |
-
from typing import List, Union, BinaryIO, Optional
|
4 |
import numpy as np
|
5 |
import time
|
6 |
import logging
|
@@ -8,6 +8,7 @@ import logging
|
|
8 |
from modules.utils.paths import DIARIZATION_MODELS_DIR
|
9 |
from modules.diarize.diarize_pipeline import DiarizationPipeline, assign_word_speakers
|
10 |
from modules.diarize.audio_loader import load_audio
|
|
|
11 |
|
12 |
|
13 |
class Diarizer:
|
@@ -23,10 +24,10 @@ class Diarizer:
|
|
23 |
|
24 |
def run(self,
|
25 |
audio: Union[str, BinaryIO, np.ndarray],
|
26 |
-
transcribed_result: List[
|
27 |
use_auth_token: str,
|
28 |
device: Optional[str] = None
|
29 |
-
):
|
30 |
"""
|
31 |
Diarize transcribed result as a post-processing
|
32 |
|
@@ -34,7 +35,7 @@ class Diarizer:
|
|
34 |
----------
|
35 |
audio: Union[str, BinaryIO, np.ndarray]
|
36 |
Audio input. This can be file path or binary type.
|
37 |
-
transcribed_result: List[
|
38 |
transcribed result through whisper.
|
39 |
use_auth_token: str
|
40 |
Huggingface token with READ permission. This is only needed the first time you download the model.
|
@@ -44,8 +45,8 @@ class Diarizer:
|
|
44 |
|
45 |
Returns
|
46 |
----------
|
47 |
-
segments_result: List[
|
48 |
-
list of
|
49 |
elapsed_time: float
|
50 |
elapsed time for running
|
51 |
"""
|
@@ -68,14 +69,21 @@ class Diarizer:
|
|
68 |
{"segments": transcribed_result}
|
69 |
)
|
70 |
|
|
|
71 |
for segment in diarized_result["segments"]:
|
|
|
72 |
speaker = "None"
|
73 |
if "speaker" in segment:
|
74 |
speaker = segment["speaker"]
|
75 |
-
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
elapsed_time = time.time() - start_time
|
78 |
-
return
|
79 |
|
80 |
def update_pipe(self,
|
81 |
use_auth_token: str,
|
|
|
1 |
import os
|
2 |
import torch
|
3 |
+
from typing import List, Union, BinaryIO, Optional, Tuple
|
4 |
import numpy as np
|
5 |
import time
|
6 |
import logging
|
|
|
8 |
from modules.utils.paths import DIARIZATION_MODELS_DIR
|
9 |
from modules.diarize.diarize_pipeline import DiarizationPipeline, assign_word_speakers
|
10 |
from modules.diarize.audio_loader import load_audio
|
11 |
+
from modules.whisper.data_classes import *
|
12 |
|
13 |
|
14 |
class Diarizer:
|
|
|
24 |
|
25 |
def run(self,
|
26 |
audio: Union[str, BinaryIO, np.ndarray],
|
27 |
+
transcribed_result: List[Segment],
|
28 |
use_auth_token: str,
|
29 |
device: Optional[str] = None
|
30 |
+
) -> Tuple[List[Segment], float]:
|
31 |
"""
|
32 |
Diarize transcribed result as a post-processing
|
33 |
|
|
|
35 |
----------
|
36 |
audio: Union[str, BinaryIO, np.ndarray]
|
37 |
Audio input. This can be file path or binary type.
|
38 |
+
transcribed_result: List[Segment]
|
39 |
transcribed result through whisper.
|
40 |
use_auth_token: str
|
41 |
Huggingface token with READ permission. This is only needed the first time you download the model.
|
|
|
45 |
|
46 |
Returns
|
47 |
----------
|
48 |
+
segments_result: List[Segment]
|
49 |
+
list of Segment that includes start, end timestamps and transcribed text
|
50 |
elapsed_time: float
|
51 |
elapsed time for running
|
52 |
"""
|
|
|
69 |
{"segments": transcribed_result}
|
70 |
)
|
71 |
|
72 |
+
segments_result = []
|
73 |
for segment in diarized_result["segments"]:
|
74 |
+
segment = segment.dict()
|
75 |
speaker = "None"
|
76 |
if "speaker" in segment:
|
77 |
speaker = segment["speaker"]
|
78 |
+
diarized_text = speaker + "|" + segment["text"].strip()
|
79 |
+
segments_result.append(Segment(
|
80 |
+
start=segment["start"],
|
81 |
+
end=segment["end"],
|
82 |
+
text=diarized_text
|
83 |
+
))
|
84 |
|
85 |
elapsed_time = time.time() - start_time
|
86 |
+
return segments_result, elapsed_time
|
87 |
|
88 |
def update_pipe(self,
|
89 |
use_auth_token: str,
|
modules/utils/subtitle_manager.py
CHANGED
@@ -1,5 +1,7 @@
|
|
1 |
import re
|
2 |
|
|
|
|
|
3 |
|
4 |
def timeformat_srt(time):
|
5 |
hours = time // 3600
|
@@ -23,6 +25,9 @@ def write_file(subtitle, output_file):
|
|
23 |
|
24 |
|
25 |
def get_srt(segments):
|
|
|
|
|
|
|
26 |
output = ""
|
27 |
for i, segment in enumerate(segments):
|
28 |
output += f"{i + 1}\n"
|
@@ -34,6 +39,9 @@ def get_srt(segments):
|
|
34 |
|
35 |
|
36 |
def get_vtt(segments):
|
|
|
|
|
|
|
37 |
output = "WEBVTT\n\n"
|
38 |
for i, segment in enumerate(segments):
|
39 |
output += f"{timeformat_vtt(segment['start'])} --> {timeformat_vtt(segment['end'])}\n"
|
@@ -44,6 +52,9 @@ def get_vtt(segments):
|
|
44 |
|
45 |
|
46 |
def get_txt(segments):
|
|
|
|
|
|
|
47 |
output = ""
|
48 |
for i, segment in enumerate(segments):
|
49 |
if segment['text'].startswith(' '):
|
|
|
1 |
import re
|
2 |
|
3 |
+
from modules.whisper.data_classes import Segment
|
4 |
+
|
5 |
|
6 |
def timeformat_srt(time):
|
7 |
hours = time // 3600
|
|
|
25 |
|
26 |
|
27 |
def get_srt(segments):
|
28 |
+
if segments and isinstance(segments[0], Segment):
|
29 |
+
segments = [seg.dict() for seg in segments]
|
30 |
+
|
31 |
output = ""
|
32 |
for i, segment in enumerate(segments):
|
33 |
output += f"{i + 1}\n"
|
|
|
39 |
|
40 |
|
41 |
def get_vtt(segments):
|
42 |
+
if segments and isinstance(segments[0], Segment):
|
43 |
+
segments = [seg.dict() for seg in segments]
|
44 |
+
|
45 |
output = "WEBVTT\n\n"
|
46 |
for i, segment in enumerate(segments):
|
47 |
output += f"{timeformat_vtt(segment['start'])} --> {timeformat_vtt(segment['end'])}\n"
|
|
|
52 |
|
53 |
|
54 |
def get_txt(segments):
|
55 |
+
if segments and isinstance(segments[0], Segment):
|
56 |
+
segments = [seg.dict() for seg in segments]
|
57 |
+
|
58 |
output = ""
|
59 |
for i, segment in enumerate(segments):
|
60 |
if segment['text'].startswith(' '):
|
modules/vad/silero_vad.py
CHANGED
@@ -5,7 +5,8 @@ import numpy as np
|
|
5 |
from typing import BinaryIO, Union, List, Optional, Tuple
|
6 |
import warnings
|
7 |
import faster_whisper
|
8 |
-
from
|
|
|
9 |
import gradio as gr
|
10 |
|
11 |
|
@@ -247,18 +248,18 @@ class SileroVAD:
|
|
247 |
|
248 |
def restore_speech_timestamps(
|
249 |
self,
|
250 |
-
segments: List[
|
251 |
speech_chunks: List[dict],
|
252 |
sampling_rate: Optional[int] = None,
|
253 |
-
) -> List[
|
254 |
if sampling_rate is None:
|
255 |
sampling_rate = self.sampling_rate
|
256 |
|
257 |
ts_map = SpeechTimestampsMap(speech_chunks, sampling_rate)
|
258 |
|
259 |
for segment in segments:
|
260 |
-
segment
|
261 |
-
segment
|
262 |
|
263 |
return segments
|
264 |
|
|
|
5 |
from typing import BinaryIO, Union, List, Optional, Tuple
|
6 |
import warnings
|
7 |
import faster_whisper
|
8 |
+
from modules.whisper.data_classes import *
|
9 |
+
from faster_whisper.transcribe import SpeechTimestampsMap
|
10 |
import gradio as gr
|
11 |
|
12 |
|
|
|
248 |
|
249 |
def restore_speech_timestamps(
|
250 |
self,
|
251 |
+
segments: List[Segment],
|
252 |
speech_chunks: List[dict],
|
253 |
sampling_rate: Optional[int] = None,
|
254 |
+
) -> List[Segment]:
|
255 |
if sampling_rate is None:
|
256 |
sampling_rate = self.sampling_rate
|
257 |
|
258 |
ts_map = SpeechTimestampsMap(speech_chunks, sampling_rate)
|
259 |
|
260 |
for segment in segments:
|
261 |
+
segment.start = ts_map.get_original_time(segment.start)
|
262 |
+
segment.start = ts_map.get_original_time(segment.start)
|
263 |
|
264 |
return segments
|
265 |
|
modules/whisper/faster_whisper_inference.py
CHANGED
@@ -40,7 +40,7 @@ class FasterWhisperInference(BaseTranscriptionPipeline):
|
|
40 |
audio: Union[str, BinaryIO, np.ndarray],
|
41 |
progress: gr.Progress = gr.Progress(),
|
42 |
*whisper_params,
|
43 |
-
) -> Tuple[List[
|
44 |
"""
|
45 |
transcribe method for faster-whisper.
|
46 |
|
@@ -55,8 +55,8 @@ class FasterWhisperInference(BaseTranscriptionPipeline):
|
|
55 |
|
56 |
Returns
|
57 |
----------
|
58 |
-
segments_result: List[
|
59 |
-
list of
|
60 |
elapsed_time: float
|
61 |
elapsed time for transcription
|
62 |
"""
|
@@ -102,11 +102,11 @@ class FasterWhisperInference(BaseTranscriptionPipeline):
|
|
102 |
segments_result = []
|
103 |
for segment in segments:
|
104 |
progress(segment.start / info.duration, desc="Transcribing..")
|
105 |
-
segments_result.append(
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
|
111 |
elapsed_time = time.time() - start_time
|
112 |
return segments_result, elapsed_time
|
|
|
40 |
audio: Union[str, BinaryIO, np.ndarray],
|
41 |
progress: gr.Progress = gr.Progress(),
|
42 |
*whisper_params,
|
43 |
+
) -> Tuple[List[Segment], float]:
|
44 |
"""
|
45 |
transcribe method for faster-whisper.
|
46 |
|
|
|
55 |
|
56 |
Returns
|
57 |
----------
|
58 |
+
segments_result: List[Segment]
|
59 |
+
list of Segment that includes start, end timestamps and transcribed text
|
60 |
elapsed_time: float
|
61 |
elapsed time for transcription
|
62 |
"""
|
|
|
102 |
segments_result = []
|
103 |
for segment in segments:
|
104 |
progress(segment.start / info.duration, desc="Transcribing..")
|
105 |
+
segments_result.append(Segment(
|
106 |
+
start=segment.start,
|
107 |
+
end=segment.end,
|
108 |
+
text=segment.text
|
109 |
+
))
|
110 |
|
111 |
elapsed_time = time.time() - start_time
|
112 |
return segments_result, elapsed_time
|
modules/whisper/insanely_fast_whisper_inference.py
CHANGED
@@ -40,7 +40,7 @@ class InsanelyFastWhisperInference(BaseTranscriptionPipeline):
|
|
40 |
audio: Union[str, np.ndarray, torch.Tensor],
|
41 |
progress: gr.Progress = gr.Progress(),
|
42 |
*whisper_params,
|
43 |
-
) -> Tuple[List[
|
44 |
"""
|
45 |
transcribe method for faster-whisper.
|
46 |
|
@@ -55,8 +55,8 @@ class InsanelyFastWhisperInference(BaseTranscriptionPipeline):
|
|
55 |
|
56 |
Returns
|
57 |
----------
|
58 |
-
segments_result: List[
|
59 |
-
list of
|
60 |
elapsed_time: float
|
61 |
elapsed time for transcription
|
62 |
"""
|
@@ -95,9 +95,17 @@ class InsanelyFastWhisperInference(BaseTranscriptionPipeline):
|
|
95 |
generate_kwargs=kwargs
|
96 |
)
|
97 |
|
98 |
-
segments_result =
|
99 |
-
|
100 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
elapsed_time = time.time() - start_time
|
102 |
return segments_result, elapsed_time
|
103 |
|
|
|
40 |
audio: Union[str, np.ndarray, torch.Tensor],
|
41 |
progress: gr.Progress = gr.Progress(),
|
42 |
*whisper_params,
|
43 |
+
) -> Tuple[List[Segment], float]:
|
44 |
"""
|
45 |
transcribe method for faster-whisper.
|
46 |
|
|
|
55 |
|
56 |
Returns
|
57 |
----------
|
58 |
+
segments_result: List[Segment]
|
59 |
+
list of Segment that includes start, end timestamps and transcribed text
|
60 |
elapsed_time: float
|
61 |
elapsed time for transcription
|
62 |
"""
|
|
|
95 |
generate_kwargs=kwargs
|
96 |
)
|
97 |
|
98 |
+
segments_result = []
|
99 |
+
for item in segments["chunks"]:
|
100 |
+
start, end = item["timestamp"][0], item["timestamp"][1]
|
101 |
+
if end is None:
|
102 |
+
end = start
|
103 |
+
segments_result.append(Segment(
|
104 |
+
text=item["text"],
|
105 |
+
start=start,
|
106 |
+
end=end
|
107 |
+
))
|
108 |
+
|
109 |
elapsed_time = time.time() - start_time
|
110 |
return segments_result, elapsed_time
|
111 |
|
modules/whisper/whisper_Inference.py
CHANGED
@@ -30,7 +30,7 @@ class WhisperInference(BaseTranscriptionPipeline):
|
|
30 |
audio: Union[str, np.ndarray, torch.Tensor],
|
31 |
progress: gr.Progress = gr.Progress(),
|
32 |
*whisper_params,
|
33 |
-
) -> Tuple[List[
|
34 |
"""
|
35 |
transcribe method for faster-whisper.
|
36 |
|
@@ -45,8 +45,8 @@ class WhisperInference(BaseTranscriptionPipeline):
|
|
45 |
|
46 |
Returns
|
47 |
----------
|
48 |
-
segments_result: List[
|
49 |
-
list of
|
50 |
elapsed_time: float
|
51 |
elapsed time for transcription
|
52 |
"""
|
@@ -59,21 +59,28 @@ class WhisperInference(BaseTranscriptionPipeline):
|
|
59 |
def progress_callback(progress_value):
|
60 |
progress(progress_value, desc="Transcribing..")
|
61 |
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
|
|
77 |
return segments_result, elapsed_time
|
78 |
|
79 |
def update_model(self,
|
|
|
30 |
audio: Union[str, np.ndarray, torch.Tensor],
|
31 |
progress: gr.Progress = gr.Progress(),
|
32 |
*whisper_params,
|
33 |
+
) -> Tuple[List[Segment], float]:
|
34 |
"""
|
35 |
transcribe method for faster-whisper.
|
36 |
|
|
|
45 |
|
46 |
Returns
|
47 |
----------
|
48 |
+
segments_result: List[Segment]
|
49 |
+
list of Segment that includes start, end timestamps and transcribed text
|
50 |
elapsed_time: float
|
51 |
elapsed time for transcription
|
52 |
"""
|
|
|
59 |
def progress_callback(progress_value):
|
60 |
progress(progress_value, desc="Transcribing..")
|
61 |
|
62 |
+
result = self.model.transcribe(audio=audio,
|
63 |
+
language=params.lang,
|
64 |
+
verbose=False,
|
65 |
+
beam_size=params.beam_size,
|
66 |
+
logprob_threshold=params.log_prob_threshold,
|
67 |
+
no_speech_threshold=params.no_speech_threshold,
|
68 |
+
task="translate" if params.is_translate and self.current_model_size in self.translatable_models else "transcribe",
|
69 |
+
fp16=True if params.compute_type == "float16" else False,
|
70 |
+
best_of=params.best_of,
|
71 |
+
patience=params.patience,
|
72 |
+
temperature=params.temperature,
|
73 |
+
compression_ratio_threshold=params.compression_ratio_threshold,
|
74 |
+
progress_callback=progress_callback,)["segments"]
|
75 |
+
segments_result = []
|
76 |
+
for segment in result:
|
77 |
+
segments_result.append(Segment(
|
78 |
+
start=segment["start"],
|
79 |
+
end=segment["end"],
|
80 |
+
text=segment["text"]
|
81 |
+
))
|
82 |
|
83 |
+
elapsed_time = time.time() - start_time
|
84 |
return segments_result, elapsed_time
|
85 |
|
86 |
def update_model(self,
|