Spaces:
Runtime error
Runtime error
Delete App_Function_Libraries/Audio_Transcription_Lib.py
Browse files
App_Function_Libraries/Audio_Transcription_Lib.py
DELETED
@@ -1,289 +0,0 @@
|
|
1 |
-
# Audio_Transcription_Lib.py
|
2 |
-
#########################################
|
3 |
-
# Transcription Library
|
4 |
-
# This library is used to perform transcription of audio files.
|
5 |
-
# Currently, uses faster_whisper for transcription.
|
6 |
-
#
|
7 |
-
####################
|
8 |
-
# Function List
|
9 |
-
#
|
10 |
-
# 1. convert_to_wav(video_file_path, offset=0, overwrite=False)
|
11 |
-
# 2. speech_to_text(audio_file_path, selected_source_lang='en', whisper_model='small.en', vad_filter=False)
|
12 |
-
#
|
13 |
-
####################
|
14 |
-
#
|
15 |
-
# Import necessary libraries to run solo for testing
|
16 |
-
import gc
|
17 |
-
import json
|
18 |
-
import logging
|
19 |
-
import os
|
20 |
-
import queue
|
21 |
-
import sys
|
22 |
-
import subprocess
|
23 |
-
import tempfile
|
24 |
-
import threading
|
25 |
-
import time
|
26 |
-
# DEBUG Imports
|
27 |
-
#from memory_profiler import profile
|
28 |
-
#import pyaudio
|
29 |
-
from faster_whisper import WhisperModel as OriginalWhisperModel
|
30 |
-
from typing import Optional, Union, List, Dict, Any
|
31 |
-
#
|
32 |
-
# Import Local
|
33 |
-
from App_Function_Libraries.Utils.Utils import load_comprehensive_config
|
34 |
-
#
|
35 |
-
#######################################################################################################################
|
36 |
-
# Function Definitions
|
37 |
-
#
|
38 |
-
|
39 |
-
# Convert video .m4a into .wav using ffmpeg
|
40 |
-
# ffmpeg -i "example.mp4" -ar 16000 -ac 1 -c:a pcm_s16le "output.wav"
|
41 |
-
# https://www.gyan.dev/ffmpeg/builds/
|
42 |
-
#
|
43 |
-
|
44 |
-
|
45 |
-
whisper_model_instance = None
|
46 |
-
config = load_comprehensive_config()
|
47 |
-
processing_choice = config.get('Processing', 'processing_choice', fallback='cpu')
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
class WhisperModel(OriginalWhisperModel):
|
52 |
-
tldw_dir = os.path.dirname(os.path.dirname(__file__))
|
53 |
-
default_download_root = os.path.join(tldw_dir, 'App_Function_Libraries', 'models', 'Whisper')
|
54 |
-
|
55 |
-
valid_model_sizes = [
|
56 |
-
"tiny.en", "tiny", "base.en", "base", "small.en", "small", "medium.en", "medium",
|
57 |
-
"large-v1", "large-v2", "large-v3", "large", "distil-large-v2", "distil-medium.en",
|
58 |
-
"distil-small.en", "distil-large-v3"
|
59 |
-
]
|
60 |
-
|
61 |
-
def __init__(
|
62 |
-
self,
|
63 |
-
model_size_or_path: str,
|
64 |
-
device: str = "auto",
|
65 |
-
device_index: Union[int, List[int]] = 0,
|
66 |
-
compute_type: str = "default",
|
67 |
-
cpu_threads: int = 16,
|
68 |
-
num_workers: int = 1,
|
69 |
-
download_root: Optional[str] = None,
|
70 |
-
local_files_only: bool = False,
|
71 |
-
files: Optional[Dict[str, Any]] = None,
|
72 |
-
**model_kwargs: Any
|
73 |
-
):
|
74 |
-
if download_root is None:
|
75 |
-
download_root = self.default_download_root
|
76 |
-
|
77 |
-
os.makedirs(download_root, exist_ok=True)
|
78 |
-
|
79 |
-
# FIXME - validate....
|
80 |
-
# Also write an integration test...
|
81 |
-
# Check if model_size_or_path is a valid model size
|
82 |
-
if model_size_or_path in self.valid_model_sizes:
|
83 |
-
# It's a model size, so we'll use the download_root
|
84 |
-
model_path = os.path.join(download_root, model_size_or_path)
|
85 |
-
if not os.path.isdir(model_path):
|
86 |
-
# If it doesn't exist, we'll let the parent class download it
|
87 |
-
model_size_or_path = model_size_or_path # Keep the original model size
|
88 |
-
else:
|
89 |
-
# If it exists, use the full path
|
90 |
-
model_size_or_path = model_path
|
91 |
-
else:
|
92 |
-
# It's not a valid model size, so assume it's a path
|
93 |
-
model_size_or_path = os.path.abspath(model_size_or_path)
|
94 |
-
|
95 |
-
super().__init__(
|
96 |
-
model_size_or_path,
|
97 |
-
device=device,
|
98 |
-
device_index=device_index,
|
99 |
-
compute_type=compute_type,
|
100 |
-
cpu_threads=cpu_threads,
|
101 |
-
num_workers=num_workers,
|
102 |
-
download_root=download_root,
|
103 |
-
local_files_only=local_files_only,
|
104 |
-
# Maybe? idk, FIXME
|
105 |
-
# files=files,
|
106 |
-
# **model_kwargs
|
107 |
-
)
|
108 |
-
|
109 |
-
def get_whisper_model(model_name, device):
|
110 |
-
global whisper_model_instance
|
111 |
-
if whisper_model_instance is None:
|
112 |
-
logging.info(f"Initializing new WhisperModel with size {model_name} on device {device}")
|
113 |
-
whisper_model_instance = WhisperModel(model_name, device=device)
|
114 |
-
return whisper_model_instance
|
115 |
-
|
116 |
-
# # FIXME: This is a temporary solution.
|
117 |
-
# # This doesn't clear older models, which means potentially a lot of memory is being used...
|
118 |
-
# def get_whisper_model(model_name, device):
|
119 |
-
# global whisper_model_instance
|
120 |
-
# if whisper_model_instance is None:
|
121 |
-
# from faster_whisper import WhisperModel
|
122 |
-
# logging.info(f"Initializing new WhisperModel with size {model_name} on device {device}")
|
123 |
-
#
|
124 |
-
# # FIXME - add logic to detect if the model is already downloaded
|
125 |
-
# # want to first check if the model is already downloaded
|
126 |
-
# # if not, download it using the existing logic in 'WhisperModel'
|
127 |
-
# # https://github.com/SYSTRAN/faster-whisper/blob/d57c5b40b06e59ec44240d93485a95799548af50/faster_whisper/transcribe.py#L584
|
128 |
-
# # Designated path should be `tldw/App_Function_Libraries/models/Whisper/`
|
129 |
-
# WhisperModel.download_root = os.path.join(os.path.dirname(__file__), 'models', 'Whisper')
|
130 |
-
# os.makedirs(WhisperModel.download_root, exist_ok=True)
|
131 |
-
# whisper_model_instance = WhisperModel(model_name, device=device)
|
132 |
-
# return whisper_model_instance
|
133 |
-
|
134 |
-
|
135 |
-
# os.system(r'.\Bin\ffmpeg.exe -ss 00:00:00 -i "{video_file_path}" -ar 16000 -ac 1 -c:a pcm_s16le "{out_path}"')
|
136 |
-
#DEBUG
|
137 |
-
#@profile
|
138 |
-
def convert_to_wav(video_file_path, offset=0, overwrite=False):
|
139 |
-
out_path = os.path.splitext(video_file_path)[0] + ".wav"
|
140 |
-
|
141 |
-
if os.path.exists(out_path) and not overwrite:
|
142 |
-
print(f"File '{out_path}' already exists. Skipping conversion.")
|
143 |
-
logging.info(f"Skipping conversion as file already exists: {out_path}")
|
144 |
-
return out_path
|
145 |
-
print("Starting conversion process of .m4a to .WAV")
|
146 |
-
out_path = os.path.splitext(video_file_path)[0] + ".wav"
|
147 |
-
|
148 |
-
try:
|
149 |
-
if os.name == "nt":
|
150 |
-
logging.debug("ffmpeg being ran on windows")
|
151 |
-
|
152 |
-
if sys.platform.startswith('win'):
|
153 |
-
ffmpeg_cmd = ".\\Bin\\ffmpeg.exe"
|
154 |
-
logging.debug(f"ffmpeg_cmd: {ffmpeg_cmd}")
|
155 |
-
else:
|
156 |
-
ffmpeg_cmd = 'ffmpeg' # Assume 'ffmpeg' is in PATH for non-Windows systems
|
157 |
-
|
158 |
-
command = [
|
159 |
-
ffmpeg_cmd, # Assuming the working directory is correctly set where .\Bin exists
|
160 |
-
"-ss", "00:00:00", # Start at the beginning of the video
|
161 |
-
"-i", video_file_path,
|
162 |
-
"-ar", "16000", # Audio sample rate
|
163 |
-
"-ac", "1", # Number of audio channels
|
164 |
-
"-c:a", "pcm_s16le", # Audio codec
|
165 |
-
out_path
|
166 |
-
]
|
167 |
-
try:
|
168 |
-
# Redirect stdin from null device to prevent ffmpeg from waiting for input
|
169 |
-
with open(os.devnull, 'rb') as null_file:
|
170 |
-
result = subprocess.run(command, stdin=null_file, text=True, capture_output=True)
|
171 |
-
if result.returncode == 0:
|
172 |
-
logging.info("FFmpeg executed successfully")
|
173 |
-
logging.debug("FFmpeg output: %s", result.stdout)
|
174 |
-
else:
|
175 |
-
logging.error("Error in running FFmpeg")
|
176 |
-
logging.error("FFmpeg stderr: %s", result.stderr)
|
177 |
-
raise RuntimeError(f"FFmpeg error: {result.stderr}")
|
178 |
-
except Exception as e:
|
179 |
-
logging.error("Error occurred - ffmpeg doesn't like windows")
|
180 |
-
raise RuntimeError("ffmpeg failed")
|
181 |
-
elif os.name == "posix":
|
182 |
-
os.system(f'ffmpeg -ss 00:00:00 -i "{video_file_path}" -ar 16000 -ac 1 -c:a pcm_s16le "{out_path}"')
|
183 |
-
else:
|
184 |
-
raise RuntimeError("Unsupported operating system")
|
185 |
-
logging.info("Conversion to WAV completed: %s", out_path)
|
186 |
-
except subprocess.CalledProcessError as e:
|
187 |
-
logging.error("Error executing FFmpeg command: %s", str(e))
|
188 |
-
raise RuntimeError("Error converting video file to WAV")
|
189 |
-
except Exception as e:
|
190 |
-
logging.error("speech-to-text: Error transcribing audio: %s", str(e))
|
191 |
-
return {"error": str(e)}
|
192 |
-
gc.collect()
|
193 |
-
return out_path
|
194 |
-
|
195 |
-
|
196 |
-
# Transcribe .wav into .segments.json
|
197 |
-
#DEBUG
|
198 |
-
#@profile
|
199 |
-
def speech_to_text(audio_file_path, selected_source_lang='en', whisper_model='medium.en', vad_filter=False, diarize=False):
|
200 |
-
global whisper_model_instance, processing_choice
|
201 |
-
logging.info('speech-to-text: Loading faster_whisper model: %s', whisper_model)
|
202 |
-
|
203 |
-
time_start = time.time()
|
204 |
-
if audio_file_path is None:
|
205 |
-
raise ValueError("speech-to-text: No audio file provided")
|
206 |
-
logging.info("speech-to-text: Audio file path: %s", audio_file_path)
|
207 |
-
|
208 |
-
try:
|
209 |
-
_, file_ending = os.path.splitext(audio_file_path)
|
210 |
-
out_file = audio_file_path.replace(file_ending, ".segments.json")
|
211 |
-
prettified_out_file = audio_file_path.replace(file_ending, ".segments_pretty.json")
|
212 |
-
if os.path.exists(out_file):
|
213 |
-
logging.info("speech-to-text: Segments file already exists: %s", out_file)
|
214 |
-
with open(out_file) as f:
|
215 |
-
global segments
|
216 |
-
segments = json.load(f)
|
217 |
-
return segments
|
218 |
-
|
219 |
-
logging.info('speech-to-text: Starting transcription...')
|
220 |
-
options = dict(language=selected_source_lang, beam_size=5, best_of=5, vad_filter=vad_filter)
|
221 |
-
transcribe_options = dict(task="transcribe", **options)
|
222 |
-
# use function and config at top of file
|
223 |
-
logging.debug("speech-to-text: Using whisper model: %s", whisper_model)
|
224 |
-
whisper_model_instance = get_whisper_model(whisper_model, processing_choice)
|
225 |
-
segments_raw, info = whisper_model_instance.transcribe(audio_file_path, **transcribe_options)
|
226 |
-
|
227 |
-
segments = []
|
228 |
-
for segment_chunk in segments_raw:
|
229 |
-
chunk = {
|
230 |
-
"Time_Start": segment_chunk.start,
|
231 |
-
"Time_End": segment_chunk.end,
|
232 |
-
"Text": segment_chunk.text
|
233 |
-
}
|
234 |
-
logging.debug("Segment: %s", chunk)
|
235 |
-
segments.append(chunk)
|
236 |
-
# Print to verify its working
|
237 |
-
print(f"{segment_chunk.start:.2f}s - {segment_chunk.end:.2f}s | {segment_chunk.text}")
|
238 |
-
|
239 |
-
# Log it as well.
|
240 |
-
logging.debug(
|
241 |
-
f"Transcribed Segment: {segment_chunk.start:.2f}s - {segment_chunk.end:.2f}s | {segment_chunk.text}")
|
242 |
-
|
243 |
-
if segments:
|
244 |
-
segments[0]["Text"] = f"This text was transcribed using whisper model: {whisper_model}\n\n" + segments[0]["Text"]
|
245 |
-
|
246 |
-
if not segments:
|
247 |
-
raise RuntimeError("No transcription produced. The audio file may be invalid or empty.")
|
248 |
-
logging.info("speech-to-text: Transcription completed in %.2f seconds", time.time() - time_start)
|
249 |
-
|
250 |
-
# Save the segments to a JSON file - prettified and non-prettified
|
251 |
-
# FIXME so this is an optional flag to save either the prettified json file or the normal one
|
252 |
-
save_json = True
|
253 |
-
if save_json:
|
254 |
-
logging.info("speech-to-text: Saving segments to JSON file")
|
255 |
-
output_data = {'segments': segments}
|
256 |
-
|
257 |
-
logging.info("speech-to-text: Saving prettified JSON to %s", prettified_out_file)
|
258 |
-
with open(prettified_out_file, 'w') as f:
|
259 |
-
json.dump(output_data, f, indent=2)
|
260 |
-
|
261 |
-
logging.info("speech-to-text: Saving JSON to %s", out_file)
|
262 |
-
with open(out_file, 'w') as f:
|
263 |
-
json.dump(output_data, f)
|
264 |
-
|
265 |
-
logging.debug(f"speech-to-text: returning {segments[:500]}")
|
266 |
-
gc.collect()
|
267 |
-
return segments
|
268 |
-
|
269 |
-
except Exception as e:
|
270 |
-
logging.error("speech-to-text: Error transcribing audio: %s", str(e))
|
271 |
-
raise RuntimeError("speech-to-text: Error transcribing audio")
|
272 |
-
|
273 |
-
|
274 |
-
def record_audio(duration, sample_rate=16000, chunk_size=1024):
|
275 |
-
pass
|
276 |
-
def save_audio_temp(audio_data, sample_rate=16000):
|
277 |
-
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as temp_file:
|
278 |
-
import wave
|
279 |
-
wf = wave.open(temp_file.name, 'wb')
|
280 |
-
wf.setnchannels(1)
|
281 |
-
wf.setsampwidth(2)
|
282 |
-
wf.setframerate(sample_rate)
|
283 |
-
wf.writeframes(audio_data)
|
284 |
-
wf.close()
|
285 |
-
return temp_file.name
|
286 |
-
|
287 |
-
#
|
288 |
-
#
|
289 |
-
#######################################################################################################################
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|