add whisper transcription

diff --git a/bin/mastering/mastering.py b/bin/mastering/mastering.py
index 63862c2675838d0f0da05d1a62e651434186482a..ae3b3b1852fffd6bdcebc353c303084a62f5042d 100755 (executable)
--- a/bin/mastering/mastering.py
+++ b/bin/mastering/mastering.py
@@ -6,14 +6,10 @@ import psutil
 import logging
 import datetime
 import argparse
-import soundfile
-import librosa
-import numpy as np
-from scipy import signal
 
 
 class Logger:
-    """A logging object"""
+    """A file logger"""
 
     def __init__(self, file):
         self.logger = logging.getLogger('myapp')
@@ -30,13 +26,14 @@ class TeleCasterMastering(object):
     raw_source_formats = ['webm',]
     source_formats = ['webm', 'mp4']
     dest_formats = {
-                   'mp3' : {'in': '', 'out': '-vn -acodec libmp3lame -aq 6'},
+                   # 'mp3' : {'in': '', 'out': '-vn -acodec libmp3lame -aq 6'},
                    'mp4' : {'in': '', 'out': '-c:v libx264-c:a aac -b:a 96k'},
                    'jpg' : {'in': '-ss 0:0:10', 'out': '-frames:v 1 -y'}
                   }
     nvidia_formats = {'mp4': {'in': '', 'out': '-c:v h264_nvenc -maxrate 1100k -c:a aac -b:a 128k'}}
     vaapi_formats = {'mp4': {'in': '-hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format vaapi',
                             'out': '-c:v h264_vaapi -c:a aac -b:a 96k'}}
+    transcription_format = "vtt"
 
     date_limit = datetime.datetime(year=2024, month=4, day=19)
     tmp_dir = "/tmp/"
@@ -67,6 +64,7 @@ class TeleCasterMastering(object):
         self.verbose_mode = args.verbose
         self.remux_only_mode = args.remux_only
         self.date_filter = args.date_filter
+        self.transcribe_mode = args.transcribe
 
         if args.input_formats:
             self.input_formats = args.input_formats
@@ -89,6 +87,9 @@ class TeleCasterMastering(object):
         return extension in exts
 
     def get_offset(self, within_file, find_file, window=10):
+        import librosa
+        import numpy as np
+        from scipy import signal
         y_within, sr_within = librosa.load(within_file, sr=None, duration=60.0)
         y_find, _ = librosa.load(find_file, sr=sr_within, duration=60.0)
         c = signal.correlate(y_within, y_find[:sr_within*window], mode='valid', method='fft')
@@ -165,6 +166,17 @@ class TeleCasterMastering(object):
 
                 self.touch(log)
 
+    def transcribe(self, file):
+        import whisper
+        filename, ext = os.path.splitext(file)
+        output_dir = os.path.dirname(file)
+        model = whisper.load_model("turbo")
+        self.logger.logger.info("Transcription started...")
+        result = model.transcribe(file, language="fr", verbose=self.verbose_mode)
+        output_writer = whisper.utils.get_writer(self.transcription_format, output_dir)
+        output_writer(result, filename)
+        self.logger.logger.info("Transcription written...")
+
     def is_processed(self, source_files):
         processed = False
         for file in source_files:
@@ -221,6 +233,8 @@ class TeleCasterMastering(object):
                                     if not file in offsets:
                                         offset = offsets[1]['offsets'][file]
                                 self.transcode(file, offset=offset)
+                            if ext == "webm" and self.transcribe_mode:
+                                self.transcribe(file)
 
 
 def main():
@@ -237,6 +251,7 @@ def main():
     parser.add_argument('-r', '--remux_only', help='remux only mode', action="store_true")
     parser.add_argument('-i','--input_formats', nargs='+', help='Required input formats')
     parser.add_argument('-o','--output_formats', nargs='+', help='Required output formats')
+    parser.add_argument('-tr', '--transcribe', help='transcribe audio to vtt', action="store_true")
 
     args = parser.parse_args()
     t = TeleCasterMastering(args)