up

README.md

@@ -4,7 +4,7 @@ emoji: 🤫
 colorFrom: indigo
 colorTo: red
 sdk: gradio
-sdk_version: 3.9.1
+sdk_version: 4.16.0
 app_file: app.py
 pinned: false
 tags:

run_demo_ct2.py

@@ -27,7 +27,7 @@ warnings.filterwarnings("ignore")
 disable_progress_bar()
 
 # DEFAULT_MODEL_NAME = "bofenghuang/whisper-large-v2-cv11-french"
-DEFAULT_MODEL_NAME = "bofenghuang/whisper-large-v2-cv11-french-ct2"
+DEFAULT_MODEL_NAME = "bofenghuang/whisper-large-v3-french"
 # CHECKPOINT_FILENAME = "checkpoint_openai.pt"
 
 GEN_KWARGS = {
@@ -35,13 +35,15 @@ GEN_KWARGS = {
     "language": "fr",
     # "without_timestamps": True,
     # decode options
-    # "beam_size": 5,
+    # "beam_size": 1,
     # "patience": 2,
     # disable fallback
     # "compression_ratio_threshold": None,
     # "logprob_threshold": None,
     # vad threshold
     # "no_speech_threshold": None,
+    # "condition_on_previous_text": False,  # todo: only for distilled version
+    "vad_filter": True,
 }
 
 logging.basicConfig(
@@ -110,7 +112,8 @@ def maybe_load_cached_pipeline(model_name):
     model = cached_models.get(model_name)
     if model is None:
         # downloaded_model_path = hf_hub_download(repo_id=model_name, filename=CHECKPOINT_FILENAME)
-        downloaded_model_path = snapshot_download(repo_id=model_name)
+        # downloaded_model_path = snapshot_download(repo_id=model_name)
+        downloaded_model_path = snapshot_download(repo_id=model_name, allow_patterns="ctranslate2/*")
 
         # model = whisper.load_model(downloaded_model_path, device=device)
         model = WhisperModel(downloaded_model_path, device=device, compute_type="float16")
@@ -233,8 +236,8 @@ with gr.Blocks() as demo:
           """
         )
 
-        microphone_input = gr.inputs.Audio(source="microphone", type="filepath", label="Record", optional=True)
-        upload_input = gr.inputs.Audio(source="upload", type="filepath", label="Upload File", optional=True)
+        microphone_input = gr.Audio(sources="microphone", type="filepath", label="Record")
+        upload_input = gr.Audio(sources="upload", type="filepath", label="Upload File")
         with_timestamps_input = gr.Checkbox(label="With timestamps?")
 
         microphone_transcribe_btn = gr.Button("Transcribe Audio")
@@ -247,10 +250,7 @@ with gr.Blocks() as demo:
         text_output_df2 = gr.DataFrame(
             value=default_text_output_df,
             label="Transcription",
-            row_count=(0, "dynamic"),
-            max_rows=10,
             wrap=True,
-            overflow_row_behaviour="paginate",
         )
 
         microphone_transcribe_btn.click(
@@ -301,7 +301,7 @@ with gr.Blocks() as demo:
           """
         )
 
-        yt_link_input = gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")
+        yt_link_input = gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")
         download_youtube_btn = gr.Button("Download Youtube video")
         downloaded_video_output = gr.Video(label="Video file", mirror_webcam=False)
         download_youtube_btn.click(download_video_from_youtube, inputs=[yt_link_input], outputs=[downloaded_video_output])
@@ -311,14 +311,10 @@ with gr.Blocks() as demo:
         text_output_df = gr.DataFrame(
             value=default_text_output_df,
             label="Transcription",
-            row_count=(0, "dynamic"),
-            max_rows=10,
             wrap=True,
-            overflow_row_behaviour="paginate",
         )
 
         video_transcribe_btn.click(video_transcribe, inputs=[downloaded_video_output, with_timestamps_input3], outputs=[text_output_df])
 
-# demo.launch(server_name="0.0.0.0", debug=True)
-# demo.launch(server_name="0.0.0.0", debug=True, share=True)
-demo.launch(enable_queue=True)
+# demo.queue(max_size=10).launch(server_name="0.0.0.0", debug=True, ssl_certfile="/home/bhuang/tools/cert.pem", ssl_keyfile="/home/bhuang/tools/key.pem", ssl_verify=False)
+demo.queue(max_size=10).launch()