feat: Improving CLI with Verbose Quantize parameters

.env.dist

@@ -19,4 +19,8 @@ ODTP_DIGITAL_TWIN=
 ODTP_EXCUTION=
 ODTP_STEP=
 ODTP_COMPONENT=
-ODTP_COMPONENT_VERSION=
+ODTP_COMPONENT_VERSION=
+
+#ODTP_API_MODE=TRUE
+#ODTP_GRADIO_SHARE=TRUE
+TODO: User and password

app/app.py

@@ -28,6 +28,8 @@ import paragraphsCreator
 from pydub import AudioSegment
 import yt_dlp
 
+from slugify import slugify
+import uuid
 
 
 
@@ -59,8 +61,8 @@ class WhisperFacade(ASRFacade):
     def __init__(self, model: str, *, quantize=False) -> None:
         print("Initialize Whisper")
         whisper_model = whisper.load_model(model)
-        if quantize:
-            print("Quantize")
+        if quantize == True:
+            print("Quantize active")
             DTYPE = torch.qint8
             qmodel: Whisper = torch.quantization.quantize_dynamic(
                 whisper_model, {torch.nn.Linear}, dtype=DTYPE
@@ -68,6 +70,7 @@ class WhisperFacade(ASRFacade):
             del whisper_model
             self.wmodel = qmodel
         else:
+            print("Quantize non active")
             self.wmodel = whisper_model
 
     def load_audio(self, file_path: str):
@@ -451,7 +454,8 @@ def convert_mpx_to_wav(file_path):
 def download_youtube_video(url, output_path='/tmp'):
     ydl_opts = {
         'format': 'bestaudio/best',
-        'outtmpl': os.path.join(output_path, '%(title)s.%(ext)s'),
+        #'outtmpl': os.path.join(output_path, '%(title)s.%(ext)s'),
+        'outtmpl': os.path.join(output_path, f"{str(uuid.uuid4())}.%(ext)s"),
         'postprocessors': [{
             'key': 'FFmpegExtractAudio',
             'preferredcodec': 'wav',
@@ -465,28 +469,29 @@ def download_youtube_video(url, output_path='/tmp'):
     with yt_dlp.YoutubeDL(ydl_opts) as ydl:
         info_dict = ydl.extract_info(url, download=True)
         output_file = ydl.prepare_filename(info_dict)
-
+        print(output_file)
         base, ext = os.path.splitext(output_file)
+        
         new_file = base + '.wav'
-        print(new_file)
         return new_file
 
 
 
 def main(args):
-    # Really dirty way to handle youtube links
-    # We should refactor this to be more robust
-    # with different input sources
-    possible_link = args.input_file.replace("/odtp/odtp-input/", "")  
-    if possible_link.startswith('http://') or possible_link.startswith('https://'):
-        file_path = download_youtube_video(possible_link, output_path=os.path.dirname(args.output_file))
+    # TODO: Take out the file_path from ODTP here
+    if args.input_file.startswith('http://') or args.input_file.startswith('https://'):
+        file_path = download_youtube_video(args.input_file, output_path=os.path.dirname(args.output_file))
+        base_slug = slugify(file_path, separator='_')
         #file_path = convert_mpx_to_wav(file_path)
     elif args.input_file.lower().endswith('.mp3'):
         file_path = convert_mpx_to_wav(args.input_file)
+        file_path = "/odtp/odtp-input/" + file_path
     elif args.input_file.lower().endswith('.wav'):
         file_path = args.input_file
+        file_path = "/odtp/odtp-input/" + file_path
     elif args.input_file.lower().endswith('.mp4'):
         file_path = convert_mpx_to_wav(args.input_file)
+        file_path = "/odtp/odtp-input/" + file_path
     else:
         raise ValueError("Input file must be an MP3, WAV or MP4 file")
     
@@ -517,7 +522,7 @@ def main(args):
         "word_timestamps": False
     }
 
-    if args.verbose == "True":
+    if args.verbose:
         print("Process diarized blocks")
 
     
@@ -527,7 +532,7 @@ def main(args):
     current_end     = None
 
     for turn, _, speaker in diarization.itertracks(yield_label=True):
-        if args.verbose=="True":
+        if args.verbose:
             print(speaker)
         if turn.end - turn.start < 0.5:
             # ignore short utterances
@@ -553,7 +558,7 @@ def main(args):
         result = asr_model.transcribe(start=start, end=end, options=whisper_options)
         language = result.get('language', args.language or 'unknown')
         
-        if args.verbose=="True":
+        if args.verbose:
             print(f"start={start:.1f}s stop={end:.1f}s lang={language} {speaker}")
 
         # Use your existing logic to write SRT, JSON, etc.
@@ -588,7 +593,7 @@ if __name__ == '__main__':
     parser.add_argument('--output-paragraphs-json-file', type=str, required=True, help="Output paragraphs file")
     parser.add_argument('--output-md-file', type=str, required=True, help="Output markdown file")
     parser.add_argument('--output-pdf-file', type=str, required=True, help="Output pdf file")
-    parser.add_argument('--verbose', type=str, required=False, help="Printing status")
+    parser.add_argument('--verbose', action='store_true', help="Printing status")
 
     args = parser.parse_args()
     main(args)

app/app.sh

@@ -3,26 +3,28 @@
 if [ -n "$LANGUAGE" ]; then
     python3 /odtp/odtp-app/app.py \
     --model $MODEL \
-    --quantize \
+    $( [ "$QUANTIZE" = "TRUE" ] && echo "--quantize" ) \
     --hf-token $HF_TOKEN \
     --task $TASK \
     --language $LANGUAGE \
-    --input-file /odtp/odtp-input/$INPUT_FILE \
+    --input-file $INPUT_FILE \
     --output-file /odtp/odtp-output/$OUTPUT_FILE.srt \
     --output-json-file /odtp/odtp-output/$OUTPUT_FILE.json \
     --output-paragraphs-json-file /odtp/odtp-output/$OUTPUT_FILE-paragraphs.json \
     --output-md-file /odtp/odtp-output/$OUTPUT_FILE.md \
-    --output-pdf-file /odtp/odtp-output/$OUTPUT_FILE.pdf
+    --output-pdf-file /odtp/odtp-output/$OUTPUT_FILE.pdf \
+    $( [ "$VERBOSE" = "TRUE" ] && echo "--verbose" )
 else
     python3 /odtp/odtp-app/app.py \
     --model $MODEL \
-    --quantize \
+    $( [ "$QUANTIZE" = "TRUE" ] && echo "--quantize" ) \
     --hf-token $HF_TOKEN \
     --task $TASK \
-    --input-file /odtp/odtp-input/$INPUT_FILE \
+    --input-file $INPUT_FILE \
     --output-file /odtp/odtp-output/$OUTPUT_FILE.srt \
     --output-json-file /odtp/odtp-output/$OUTPUT_FILE.json \
     --output-paragraphs-json-file /odtp/odtp-output/$OUTPUT_FILE-paragraphs.json \
     --output-md-file /odtp/odtp-output/$OUTPUT_FILE.md \
-    --output-pdf-file /odtp/odtp-output/$OUTPUT_FILE.pdf
+    --output-pdf-file /odtp/odtp-output/$OUTPUT_FILE.pdf \
+    $( [ "$VERBOSE" = "TRUE" ] && echo "--verbose" )
 fi

app/gradio_app.py

@@ -5,6 +5,8 @@ import shutil
 import subprocess
 import threading
 import time
+import argparse
+import uuid
 
 def create_temp_structure():
     """Create temporary ODTP folder structure"""
@@ -22,7 +24,13 @@ def cleanup_temp(temp_dir):
     """Remove temporary folder structure"""
     shutil.rmtree(temp_dir)
 
-def process_audio(audio_file, model, task, language, hf_token=None):
+def process_audio(audio_file, 
+                  youtube_link, 
+                  model, 
+                  task, 
+                  language, 
+                  hf_token=None,
+                  quantize=False):
     """Process audio file with Whisper and Pyannote"""
     # Create temp structure
     temp_dir = create_temp_structure()
@@ -30,16 +38,28 @@ def process_audio(audio_file, model, task, language, hf_token=None):
     start_time = time.time()
     print(f"Processing started at: {time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(start_time))}")
     
-    # Copy input file
-    input_path = os.path.join(temp_dir, "odtp-input", "input.wav")
-    shutil.copy2(audio_file, input_path)
-
-    # Prepare output paths #TODO: Add uuid to output file names
-    output_base = audio_file.split("/")[-1].replace(".wav", "")
-    output_srt = os.path.join(temp_dir, "odtp-output", #temp_dir
-        f"{output_base}_{task}.srt")
-    output_json = os.path.join(temp_dir, "odtp-output",
-        f"{output_base}_{task}.json")
+    if youtube_link:
+        input_path = youtube_link
+
+        id = str(uuid.uuid4())
+        output_base = f"{id}"
+        output_srt = os.path.join(temp_dir, "odtp-output", f"{id}.srt")
+        output_json = os.path.join(temp_dir, "odtp-output", f"{id}.json")
+        output_paragraphs_json = os.path.join(temp_dir, "odtp-output", f"{id}_paragraphs.json")
+        output_md_file = os.path.join(temp_dir, "odtp-output", f"{id}.md")
+        output_pdf_file = os.path.join(temp_dir, "odtp-output", f"{id}.pdf")
+    
+    else:
+        # Copy input file
+        input_path = os.path.join(temp_dir, "odtp-input", "input.wav")
+        shutil.copy2(audio_file, input_path)
+
+        output_base = audio_file.split("/")[-1].replace(".wav", "")
+        output_srt = os.path.join(temp_dir, "odtp-output", f"{output_base}_{task}.srt")
+        output_json = os.path.join(temp_dir, "odtp-output", f"{output_base}_{task}.json")
+        output_paragraphs_json = os.path.join(temp_dir, "odtp-output", f"{output_base}_{task}_paragraphs.json")
+        output_md_file = os.path.join(temp_dir, "odtp-output", f"{output_base}_{task}.md")
+        output_pdf_file = os.path.join(temp_dir, "odtp-output", f"{output_base}_{task}.pdf")
     
     # Use HF_TOKEN from environment if not provided
     if not hf_token:
@@ -51,14 +71,19 @@ def process_audio(audio_file, model, task, language, hf_token=None):
     cmd = [
         "python3", "/odtp/odtp-app/app.py",
         "--model", model,
-        "--quantize",
         "--hf-token", hf_token,
         "--task", task,
         "--input-file", input_path,
         "--output-file", output_srt,
         "--output-json-file", output_json,
+        "--output-paragraphs-json-file", output_paragraphs_json,
+        "--output-md-file", output_md_file,
+        "--output-pdf-file", output_pdf_file,
         "--verbose", "False"
     ]
+
+    if quantize:
+        cmd.append("--quantize")
     
     if language != "auto":
         cmd.extend(["--language", language])
@@ -84,7 +109,7 @@ def process_audio(audio_file, model, task, language, hf_token=None):
     total_duration_str = f"{int(hours)}h {int(minutes)}m {int(seconds)}s"
     print(f"Total processing time: {total_duration_str}")
 
-    return total_duration_str, srt_content, json_content, output_srt, output_json
+    return total_duration_str, srt_content, json_content, output_srt, output_json, output_paragraphs_json, output_md_file, output_pdf_file
 
 # Define Gradio interface
 with gr.Blocks() as demo:
@@ -97,10 +122,23 @@ with gr.Blocks() as demo:
                 label="Upload Audio File (WAV format)"
             )
             model = gr.Dropdown(
-                choices=["tiny", "base", "small", "medium", "large", "large-v2", "large-v3", "large-v3-turbo", "softcatala/whisper-base-ca", "projecte-aina/whisper-large-v3-ca-3catparla"],
+                choices=["tiny", 
+                         "base", 
+                         "small", 
+                         "medium", 
+                         "large", 
+                         "large-v2", 
+                         "large-v3", 
+                         "large-v3-turbo", 
+                         "softcatala/whisper-base-ca", 
+                         "projecte-aina/whisper-large-v3-ca-3catparla"],
                 value="base",
                 label="Whisper Model"
             )
+            youtube_link = gr.Textbox(
+                type="text",
+                label="Youtube Link (optional)",
+            )
             task = gr.Dropdown(
                 choices=["transcribe", "translate"],
                 value="transcribe",
@@ -116,6 +154,11 @@ with gr.Blocks() as demo:
                 type="password",
                 placeholder="Leave blank if not applicable"
             )
+            quantize = gr.Checkbox(
+                label="Quantize Model",
+                value=False
+            )
+
             submit_btn = gr.Button("Process Audio")
             
         with gr.Column():
@@ -139,14 +182,39 @@ with gr.Blocks() as demo:
                 label="Download JSON File",
                 type="binary"
             )
+            json_paragraphs_download = gr.File(
+                label="Download JSON Paragraphs File",
+                type="binary"
+            )
+            md_download = gr.File(
+                label="Download MD File",
+                type="binary"
+            )
+            pdf_download = gr.File(
+                label="Download PDF File",
+                type="binary"
+            )
     
     submit_btn.click(
         fn=process_audio,
-        inputs=[audio_input, model, task, language, hf_token],
-        outputs=[information, srt_output, json_output, srt_download, json_download]
+        inputs=[audio_input, 
+                youtube_link, 
+                model, 
+                task, 
+                language, 
+                hf_token,
+                quantize],
+        outputs=[information, 
+                 srt_output, 
+                 json_output, 
+                 srt_download, 
+                 json_download, 
+                 json_paragraphs_download,
+                 md_download,
+                 pdf_download]
     )
 
-import argparse
+
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Launch Gradio app with optional sharing.")

requirements.txt

@@ -10,4 +10,5 @@ gradio==5.5.0
 numpy==1.24.4
 md2pdf==1.0.1
 transformers==4.48.0
-yt-dlp
+yt-dlp
+python-slugify