Spaces:
Running
Running
File size: 8,293 Bytes
0c30c9f ed28876 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 |
# Audio_Transcription_Lib.py
#########################################
# Transcription Library
# This library is used to perform transcription of audio files.
# Currently, uses faster_whisper for transcription.
#
####################
# Function List
#
# 1. convert_to_wav(video_file_path, offset=0, overwrite=False)
# 2. speech_to_text(audio_file_path, selected_source_lang='en', whisper_model='small.en', vad_filter=False)
#
####################
#
# Import necessary libraries to run solo for testing
import gc
import json
import logging
import os
import queue
import sys
import subprocess
import tempfile
import threading
import time
import configparser
# DEBUG Imports
#from memory_profiler import profile
#import pyaudio
from App_Function_Libraries.Utils.Utils import load_comprehensive_config
# Import Local
#
#######################################################################################################################
# Function Definitions
#
# Convert video .m4a into .wav using ffmpeg
# ffmpeg -i "example.mp4" -ar 16000 -ac 1 -c:a pcm_s16le "output.wav"
# https://www.gyan.dev/ffmpeg/builds/
#
whisper_model_instance = None
config = load_comprehensive_config()
processing_choice = config.get('Processing', 'processing_choice', fallback='cpu')
# FIXME: This is a temporary solution.
# This doesn't clear older models, which means potentially a lot of memory is being used...
def get_whisper_model(model_name, device):
global whisper_model_instance
if whisper_model_instance is None:
from faster_whisper import WhisperModel
logging.info(f"Initializing new WhisperModel with size {model_name} on device {device}")
whisper_model_instance = WhisperModel(model_name, device=device)
return whisper_model_instance
# os.system(r'.\Bin\ffmpeg.exe -ss 00:00:00 -i "{video_file_path}" -ar 16000 -ac 1 -c:a pcm_s16le "{out_path}"')
#DEBUG
#@profile
def convert_to_wav(video_file_path, offset=0, overwrite=False):
out_path = os.path.splitext(video_file_path)[0] + ".wav"
if os.path.exists(out_path) and not overwrite:
print(f"File '{out_path}' already exists. Skipping conversion.")
logging.info(f"Skipping conversion as file already exists: {out_path}")
return out_path
print("Starting conversion process of .m4a to .WAV")
out_path = os.path.splitext(video_file_path)[0] + ".wav"
try:
if os.name == "nt":
logging.debug("ffmpeg being ran on windows")
if sys.platform.startswith('win'):
ffmpeg_cmd = ".\\Bin\\ffmpeg.exe"
logging.debug(f"ffmpeg_cmd: {ffmpeg_cmd}")
else:
ffmpeg_cmd = 'ffmpeg' # Assume 'ffmpeg' is in PATH for non-Windows systems
command = [
ffmpeg_cmd, # Assuming the working directory is correctly set where .\Bin exists
"-ss", "00:00:00", # Start at the beginning of the video
"-i", video_file_path,
"-ar", "16000", # Audio sample rate
"-ac", "1", # Number of audio channels
"-c:a", "pcm_s16le", # Audio codec
out_path
]
try:
# Redirect stdin from null device to prevent ffmpeg from waiting for input
with open(os.devnull, 'rb') as null_file:
result = subprocess.run(command, stdin=null_file, text=True, capture_output=True)
if result.returncode == 0:
logging.info("FFmpeg executed successfully")
logging.debug("FFmpeg output: %s", result.stdout)
else:
logging.error("Error in running FFmpeg")
logging.error("FFmpeg stderr: %s", result.stderr)
raise RuntimeError(f"FFmpeg error: {result.stderr}")
except Exception as e:
logging.error("Error occurred - ffmpeg doesn't like windows")
raise RuntimeError("ffmpeg failed")
elif os.name == "posix":
os.system(f'ffmpeg -ss 00:00:00 -i "{video_file_path}" -ar 16000 -ac 1 -c:a pcm_s16le "{out_path}"')
else:
raise RuntimeError("Unsupported operating system")
logging.info("Conversion to WAV completed: %s", out_path)
except subprocess.CalledProcessError as e:
logging.error("Error executing FFmpeg command: %s", str(e))
raise RuntimeError("Error converting video file to WAV")
except Exception as e:
logging.error("speech-to-text: Error transcribing audio: %s", str(e))
return {"error": str(e)}
gc.collect()
return out_path
# Transcribe .wav into .segments.json
#DEBUG
#@profile
def speech_to_text(audio_file_path, selected_source_lang='en', whisper_model='medium.en', vad_filter=False, diarize=False):
global whisper_model_instance, processing_choice
logging.info('speech-to-text: Loading faster_whisper model: %s', whisper_model)
time_start = time.time()
if audio_file_path is None:
raise ValueError("speech-to-text: No audio file provided")
logging.info("speech-to-text: Audio file path: %s", audio_file_path)
try:
_, file_ending = os.path.splitext(audio_file_path)
out_file = audio_file_path.replace(file_ending, ".segments.json")
prettified_out_file = audio_file_path.replace(file_ending, ".segments_pretty.json")
if os.path.exists(out_file):
logging.info("speech-to-text: Segments file already exists: %s", out_file)
with open(out_file) as f:
global segments
segments = json.load(f)
return segments
logging.info('speech-to-text: Starting transcription...')
options = dict(language=selected_source_lang, beam_size=5, best_of=5, vad_filter=vad_filter)
transcribe_options = dict(task="transcribe", **options)
# use function and config at top of file
whisper_model_instance = get_whisper_model(whisper_model, processing_choice)
segments_raw, info = whisper_model_instance.transcribe(audio_file_path, **transcribe_options)
segments = []
for segment_chunk in segments_raw:
chunk = {
"Time_Start": segment_chunk.start,
"Time_End": segment_chunk.end,
"Text": segment_chunk.text
}
logging.debug("Segment: %s", chunk)
segments.append(chunk)
# Print to verify its working
print(f"{segment_chunk.start:.2f}s - {segment_chunk.end:.2f}s | {segment_chunk.text}")
# Log it as well.
logging.debug(
f"Transcribed Segment: {segment_chunk.start:.2f}s - {segment_chunk.end:.2f}s | {segment_chunk.text}")
if segments:
segments[0]["Text"] = f"This text was transcribed using whisper model: {whisper_model}\n\n" + segments[0]["Text"]
if not segments:
raise RuntimeError("No transcription produced. The audio file may be invalid or empty.")
logging.info("speech-to-text: Transcription completed in %.2f seconds", time.time() - time_start)
# Save the segments to a JSON file - prettified and non-prettified
# FIXME so this is an optional flag to save either the prettified json file or the normal one
save_json = True
if save_json:
logging.info("speech-to-text: Saving segments to JSON file")
output_data = {'segments': segments}
logging.info("speech-to-text: Saving prettified JSON to %s", prettified_out_file)
with open(prettified_out_file, 'w') as f:
json.dump(output_data, f, indent=2)
logging.info("speech-to-text: Saving JSON to %s", out_file)
with open(out_file, 'w') as f:
json.dump(output_data, f)
logging.debug(f"speech-to-text: returning {segments[:500]}")
gc.collect()
return segments
except Exception as e:
logging.error("speech-to-text: Error transcribing audio: %s", str(e))
raise RuntimeError("speech-to-text: Error transcribing audio")
#
#
####################################################################################################################### |