Caption daemon

1 files changed, 25 insertions, 21 deletions

diff --git a/roles/caption/templates/process-captions.py b/roles/caption/templates/process-captions.py
index 6ad890a..72e9ad2 100755
--- a/roles/caption/templates/process-captions.py
+++ b/roles/caption/templates/process-captions.py
@@ -46,7 +46,8 @@ AUDIO_REGEXP = '\.(ogg|opus)$'
 ALWAYS = False
 TRIM_AUDIO = False
 MODEL = os.environ.get('MODEL', 'large')  # Set to tiny for testing
-JSON_FILE = '/data/emacsconf/2022/talks.json'
+WORK_DIR = "{{ emacsconf_caption_dir }}"
+JSON_FILE = os.path.join(WORK_DIR, 'talks.json')
 
 def get_slug_from_filename(filename):
     m = re.search('emacsconf-[0-9]+-([a-z]+)--', filename)
@@ -100,14 +101,14 @@ def get_files_to_work_on(directory):
     return needs_work
 
 def extract_audio(work):
-    output = subprocess.check_output(['ffprobe', video_file], stderr=subprocess.STDOUT)
+    output = subprocess.check_output(['ffprobe', work['video']], stderr=subprocess.STDOUT)
     extension = 'opus'
     if 'Audio: vorbis' in output.decode():
         extension = 'ogg'
-    new_file = os.path.join(os.path.dirname(video_file), base_name(video_file) + '.' + extension)
-    acodec = 'copy' if re.search('webm$', video_file) else 'libopus'
-    log("Extracting audio from %s acodec %s" % (video_file, acodec))
-    output = subprocess.check_output(['ffmpeg', '-y', '-i', video_file, '-acodec', acodec, '-vn', new_file], stderr=subprocess.STDOUT)
+    new_file = work['base'] + '.' + extension
+    acodec = 'copy' if re.search('webm$', work['video']) else 'libopus'
+    log("Extracting audio from %s acodec %s" % (work['video'], acodec))
+    output = subprocess.check_output(['ffmpeg', '-y', '-i', work['video'], '-acodec', acodec, '-vn', new_file], stderr=subprocess.STDOUT)
     work['audio'] = new_file
     return work
 
@@ -213,19 +214,22 @@ def base_name(s):
 # assert(base_name('/home/sachac/current/sqlite/emacsconf-2022-sqlite--using-sqlite-as-a-data-source-a-framework-and-an-example--andrew-hyatt--normalized.webm.vtt') == 'emacsconf-2022-sqlite--using-sqlite-as-a-data-source-a-framework-and-an-example--andrew-hyatt--main')
 
 log(f"MODEL {MODEL} ALWAYS {ALWAYS} TRIM_AUDIO {TRIM_AUDIO}")
-directory = sys.argv[1] if len(sys.argv) > 1 else "~/current"
+directory = sys.argv[1] if len(sys.argv) > 1 else WORK_DIR
+
 needs_work = get_files_to_work_on(directory)
-if THREADS > 0:
-    torch.set_num_threads(THREADS)
-for work in needs_work:
-    log("Started processing %s" % work['base'])
-    if work['audio']:
-        if ALWAYS or not 'vtt' in work:
-            work = generate_captions(work)
-        if ALWAYS or not 'srv2' in work:
-            work = generate_srv2(work)
-#     print("Aligning words", audio_file, datetime.datetime.now())
-#     word_cuts = align_words(cuts)
-#     convert_cuts_to_word_timing(audio_file, word_cuts)
-    log("Done %s" % str(work['base']))
-    
+if len(needs_work) > 0:
+    if THREADS > 0:
+        torch.set_num_threads(THREADS)
+    for work in needs_work:
+        log("Started processing %s" % work['base'])
+        if work['audio']:
+            if ALWAYS or not 'vtt' in work:
+                work = generate_captions(work)
+            if ALWAYS or not 'srv2' in work:
+                work = generate_srv2(work)
+    #     print("Aligning words", audio_file, datetime.datetime.now())
+    #     word_cuts = align_words(cuts)
+    #     convert_cuts_to_word_timing(audio_file, word_cuts)
+        log("Done %s" % str(work['base']))
+else:
+    log("No work needed.")