Update modules/whisper/faster_whisper_inference.py
Browse files
modules/whisper/faster_whisper_inference.py
CHANGED
@@ -1,5 +1,6 @@
|
|
1 |
import os
|
2 |
import time
|
|
|
3 |
import numpy as np
|
4 |
import torch
|
5 |
from typing import BinaryIO, Union, Tuple, List
|
@@ -12,11 +13,11 @@ import gradio as gr
|
|
12 |
from argparse import Namespace
|
13 |
|
14 |
from modules.utils.paths import (FASTER_WHISPER_MODELS_DIR, DIARIZATION_MODELS_DIR, UVR_MODELS_DIR, OUTPUT_DIR)
|
15 |
-
from modules.whisper.
|
16 |
-
from modules.whisper.
|
17 |
|
18 |
|
19 |
-
class FasterWhisperInference(
|
20 |
def __init__(self,
|
21 |
model_dir: str = FASTER_WHISPER_MODELS_DIR,
|
22 |
diarization_model_dir: str = DIARIZATION_MODELS_DIR,
|
@@ -35,14 +36,12 @@ class FasterWhisperInference(WhisperBase):
|
|
35 |
self.model_paths = self.get_model_paths()
|
36 |
self.device = self.get_device()
|
37 |
self.available_models = self.model_paths.keys()
|
38 |
-
self.available_compute_types = ctranslate2.get_supported_compute_types(
|
39 |
-
"cuda") if self.device == "cuda" else ctranslate2.get_supported_compute_types("cpu")
|
40 |
|
41 |
def transcribe(self,
|
42 |
audio: Union[str, BinaryIO, np.ndarray],
|
43 |
progress: gr.Progress = gr.Progress(),
|
44 |
*whisper_params,
|
45 |
-
) -> Tuple[List[
|
46 |
"""
|
47 |
transcribe method for faster-whisper.
|
48 |
|
@@ -57,32 +56,22 @@ class FasterWhisperInference(WhisperBase):
|
|
57 |
|
58 |
Returns
|
59 |
----------
|
60 |
-
segments_result: List[
|
61 |
-
list of
|
62 |
elapsed_time: float
|
63 |
elapsed time for transcription
|
64 |
"""
|
65 |
start_time = time.time()
|
66 |
|
67 |
-
params =
|
68 |
|
69 |
if params.model_size != self.current_model_size or self.model is None or self.current_compute_type != params.compute_type:
|
70 |
self.update_model(params.model_size, params.compute_type, progress)
|
71 |
|
72 |
-
# None parameters with Textboxes: https://github.com/gradio-app/gradio/issues/8723
|
73 |
-
if not params.initial_prompt:
|
74 |
-
params.initial_prompt = None
|
75 |
-
if not params.prefix:
|
76 |
-
params.prefix = None
|
77 |
-
if not params.hotwords:
|
78 |
-
params.hotwords = None
|
79 |
-
|
80 |
-
params.suppress_tokens = self.format_suppress_tokens_str(params.suppress_tokens)
|
81 |
-
|
82 |
segments, info = self.model.transcribe(
|
83 |
audio=audio,
|
84 |
language=params.lang,
|
85 |
-
task="translate" if params.is_translate
|
86 |
beam_size=params.beam_size,
|
87 |
log_prob_threshold=params.log_prob_threshold,
|
88 |
no_speech_threshold=params.no_speech_threshold,
|
@@ -109,16 +98,12 @@ class FasterWhisperInference(WhisperBase):
|
|
109 |
language_detection_segments=params.language_detection_segments,
|
110 |
prompt_reset_on_temperature=params.prompt_reset_on_temperature,
|
111 |
)
|
112 |
-
progress(0, desc="Loading audio
|
113 |
|
114 |
segments_result = []
|
115 |
for segment in segments:
|
116 |
-
progress(segment.start / info.duration, desc="Transcribing
|
117 |
-
segments_result.append(
|
118 |
-
"start": segment.start,
|
119 |
-
"end": segment.end,
|
120 |
-
"text": segment.text
|
121 |
-
})
|
122 |
|
123 |
elapsed_time = time.time() - start_time
|
124 |
return segments_result, elapsed_time
|
@@ -134,21 +119,43 @@ class FasterWhisperInference(WhisperBase):
|
|
134 |
Parameters
|
135 |
----------
|
136 |
model_size: str
|
137 |
-
Size of whisper model
|
|
|
138 |
compute_type: str
|
139 |
Compute type for transcription.
|
140 |
see more info : https://opennmt.net/CTranslate2/quantization.html
|
141 |
progress: gr.Progress
|
142 |
Indicator to show progress directly in gradio.
|
143 |
"""
|
144 |
-
progress(0, desc="Initializing Model
|
145 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
self.current_compute_type = compute_type
|
147 |
self.model = faster_whisper.WhisperModel(
|
148 |
device=self.device,
|
149 |
model_size_or_path=self.current_model_size,
|
150 |
download_root=self.model_dir,
|
151 |
-
compute_type=self.current_compute_type
|
|
|
152 |
)
|
153 |
|
154 |
def get_model_paths(self):
|
@@ -163,7 +170,7 @@ class FasterWhisperInference(WhisperBase):
|
|
163 |
faster_whisper_prefix = "models--Systran--faster-whisper-"
|
164 |
|
165 |
existing_models = os.listdir(self.model_dir)
|
166 |
-
wrong_dirs = [".locks"]
|
167 |
existing_models = list(set(existing_models) - set(wrong_dirs))
|
168 |
|
169 |
for model_name in existing_models:
|
@@ -189,4 +196,4 @@ class FasterWhisperInference(WhisperBase):
|
|
189 |
raise ValueError("Invalid Suppress Tokens. The value must be type of List[int]")
|
190 |
return suppress_tokens
|
191 |
except Exception as e:
|
192 |
-
raise ValueError("Invalid Suppress Tokens. The value must be type of List[int]")
|
|
|
1 |
import os
|
2 |
import time
|
3 |
+
import huggingface_hub
|
4 |
import numpy as np
|
5 |
import torch
|
6 |
from typing import BinaryIO, Union, Tuple, List
|
|
|
13 |
from argparse import Namespace
|
14 |
|
15 |
from modules.utils.paths import (FASTER_WHISPER_MODELS_DIR, DIARIZATION_MODELS_DIR, UVR_MODELS_DIR, OUTPUT_DIR)
|
16 |
+
from modules.whisper.data_classes import *
|
17 |
+
from modules.whisper.base_transcription_pipeline import BaseTranscriptionPipeline
|
18 |
|
19 |
|
20 |
+
class FasterWhisperInference(BaseTranscriptionPipeline):
|
21 |
def __init__(self,
|
22 |
model_dir: str = FASTER_WHISPER_MODELS_DIR,
|
23 |
diarization_model_dir: str = DIARIZATION_MODELS_DIR,
|
|
|
36 |
self.model_paths = self.get_model_paths()
|
37 |
self.device = self.get_device()
|
38 |
self.available_models = self.model_paths.keys()
|
|
|
|
|
39 |
|
40 |
def transcribe(self,
|
41 |
audio: Union[str, BinaryIO, np.ndarray],
|
42 |
progress: gr.Progress = gr.Progress(),
|
43 |
*whisper_params,
|
44 |
+
) -> Tuple[List[Segment], float]:
|
45 |
"""
|
46 |
transcribe method for faster-whisper.
|
47 |
|
|
|
56 |
|
57 |
Returns
|
58 |
----------
|
59 |
+
segments_result: List[Segment]
|
60 |
+
list of Segment that includes start, end timestamps and transcribed text
|
61 |
elapsed_time: float
|
62 |
elapsed time for transcription
|
63 |
"""
|
64 |
start_time = time.time()
|
65 |
|
66 |
+
params = WhisperParams.from_list(list(whisper_params))
|
67 |
|
68 |
if params.model_size != self.current_model_size or self.model is None or self.current_compute_type != params.compute_type:
|
69 |
self.update_model(params.model_size, params.compute_type, progress)
|
70 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
segments, info = self.model.transcribe(
|
72 |
audio=audio,
|
73 |
language=params.lang,
|
74 |
+
task="translate" if params.is_translate else "transcribe",
|
75 |
beam_size=params.beam_size,
|
76 |
log_prob_threshold=params.log_prob_threshold,
|
77 |
no_speech_threshold=params.no_speech_threshold,
|
|
|
98 |
language_detection_segments=params.language_detection_segments,
|
99 |
prompt_reset_on_temperature=params.prompt_reset_on_temperature,
|
100 |
)
|
101 |
+
progress(0, desc="Loading audio..")
|
102 |
|
103 |
segments_result = []
|
104 |
for segment in segments:
|
105 |
+
progress(segment.start / info.duration, desc="Transcribing..")
|
106 |
+
segments_result.append(Segment.from_faster_whisper(segment))
|
|
|
|
|
|
|
|
|
107 |
|
108 |
elapsed_time = time.time() - start_time
|
109 |
return segments_result, elapsed_time
|
|
|
119 |
Parameters
|
120 |
----------
|
121 |
model_size: str
|
122 |
+
Size of whisper model. If you enter the huggingface repo id, it will try to download the model
|
123 |
+
automatically from huggingface.
|
124 |
compute_type: str
|
125 |
Compute type for transcription.
|
126 |
see more info : https://opennmt.net/CTranslate2/quantization.html
|
127 |
progress: gr.Progress
|
128 |
Indicator to show progress directly in gradio.
|
129 |
"""
|
130 |
+
progress(0, desc="Initializing Model..")
|
131 |
+
|
132 |
+
model_size_dirname = model_size.replace("/", "--") if "/" in model_size else model_size
|
133 |
+
if model_size not in self.model_paths and model_size_dirname not in self.model_paths:
|
134 |
+
print(f"Model is not detected. Trying to download \"{model_size}\" from huggingface to "
|
135 |
+
f"\"{os.path.join(self.model_dir, model_size_dirname)} ...")
|
136 |
+
huggingface_hub.snapshot_download(
|
137 |
+
model_size,
|
138 |
+
local_dir=os.path.join(self.model_dir, model_size_dirname),
|
139 |
+
)
|
140 |
+
self.model_paths = self.get_model_paths()
|
141 |
+
gr.Info(f"Model is downloaded with the name \"{model_size_dirname}\"")
|
142 |
+
|
143 |
+
self.current_model_size = self.model_paths[model_size_dirname]
|
144 |
+
|
145 |
+
local_files_only = False
|
146 |
+
hf_prefix = "models--Systran--faster-whisper-"
|
147 |
+
official_model_path = os.path.join(self.model_dir, hf_prefix+model_size)
|
148 |
+
if ((os.path.isdir(self.current_model_size) and os.path.exists(self.current_model_size)) or
|
149 |
+
(model_size in faster_whisper.available_models() and os.path.exists(official_model_path))):
|
150 |
+
local_files_only = True
|
151 |
+
|
152 |
self.current_compute_type = compute_type
|
153 |
self.model = faster_whisper.WhisperModel(
|
154 |
device=self.device,
|
155 |
model_size_or_path=self.current_model_size,
|
156 |
download_root=self.model_dir,
|
157 |
+
compute_type=self.current_compute_type,
|
158 |
+
local_files_only=local_files_only
|
159 |
)
|
160 |
|
161 |
def get_model_paths(self):
|
|
|
170 |
faster_whisper_prefix = "models--Systran--faster-whisper-"
|
171 |
|
172 |
existing_models = os.listdir(self.model_dir)
|
173 |
+
wrong_dirs = [".locks", "faster_whisper_models_will_be_saved_here"]
|
174 |
existing_models = list(set(existing_models) - set(wrong_dirs))
|
175 |
|
176 |
for model_name in existing_models:
|
|
|
196 |
raise ValueError("Invalid Suppress Tokens. The value must be type of List[int]")
|
197 |
return suppress_tokens
|
198 |
except Exception as e:
|
199 |
+
raise ValueError("Invalid Suppress Tokens. The value must be type of List[int]")
|