fix: try again

app.py

@@ -0,0 +1,141 @@
+import whisper
+import gradio as gr
+import ffmpeg
+import youtube_dl
+import os
+
+youtube_livestream_codes = [
+    91,
+    92,
+    93,
+    94,
+    95,
+    96,
+    300,
+    301,
+]
+youtube_mp4_codes = [
+    298,
+    18,
+    22,
+    140,
+    133,
+    134
+]
+
+import sys
+
+def get_video_metadata(video_url: str = "https://www.youtube.com/watch?v=21X5lGlDOfg&ab_channel=NASA")-> dict:
+    with youtube_dl.YoutubeDL({'outtmpl': '%(id)s.%(ext)s'}) as ydl:
+        info_dict = ydl.extract_info(video_url, download=False)
+        video_title = info_dict.get('title', None)
+        uploader_id = info_dict.get('uploader_id', None)
+        print(f"[youtube] {video_title}: {uploader_id}")
+    return info_dict
+
+
+def parse_metadata(metadata) -> dict:
+    """
+    Parse metadata and send to discord.
+    After a video is done recording, 
+    it will have both the livestream format and the mp4 format.
+    """
+    # send metadata to discord
+    formats = metadata.get("formats", [])
+    # filter for ext = mp4
+    mp4_formats = [f for f in formats if f.get("ext", "") == "mp4"]
+    format_ids = [int(f.get("format_id", 0)) for f in mp4_formats]
+    video_entries = sorted(set(format_ids).intersection(youtube_mp4_codes))
+
+    is_livestream = True
+    if len(video_entries) > 0:
+        # use video format id over livestream id if available
+        selected_id = video_entries[0]
+        is_livestream = False
+
+
+    return {
+        "selected_id": selected_id,
+        "is_livestream": is_livestream,
+    }
+
+def get_video(url: str, config: dict):
+    """
+    Get video from start time.
+    """
+    # result = subprocess.run()
+    # could delay start time by a few seconds to just sync up and capture the full video length
+    # but would need to time how long it takes to fetch the video using youtube-dl and other adjustments and start a bit before
+    filename = config.get("filename", "livestream01.mp4")
+    end = config.get("end", "00:15:00")
+    overlay_file = ffmpeg.input(filename)
+    (
+        ffmpeg
+        .input(url, t=end)
+        .output(filename)
+        .run()
+    )
+
+def get_all_files(url: str, end: str = "00:15:00"):
+    metadata = get_video_metadata(url)
+    temp_dict = parse_metadata(metadata)
+    selected_id = temp_dict.get("selected_id", 0)
+    formats = metadata.get("formats", [])
+    selected_format = [f for f in formats if f.get("format_id", "") == str(selected_id)][0]
+    format_url = selected_format.get("url", "")
+    filename = "temp.mp4"
+    get_video(format_url, {"filename": filename, "end": end})
+    return filename
+
+def get_text_from_mp3_whisper(inputType:str, mp3_file: str, url_path: str, taskName: str, srcLanguage: str)->str:
+    model = whisper.load_model("medium")
+    # options = whisper.DecodingOptions(language="en", without_timestamps=True)
+    options = dict(language=srcLanguage)
+    transcribe_options = dict(task=taskName, **options)
+    if inputType == "url":
+        filename = get_all_files(url_path)
+        result = model.transcribe(filename, **transcribe_options)
+    else:
+        result = model.transcribe(mp3_file, **transcribe_options)
+    # adjust for spacy mode
+    html_text = ""
+    lines = []
+    for count, segment in enumerate(result.get("segments")):
+        # print(segment)
+        start = segment.get("start")
+        end = segment.get("end")
+        lines.append(f"{count}")
+        lines.append(f"{second_to_timecode(start)} --> {second_to_timecode(end)}")
+        lines.append(segment.get("text", "").strip())
+        lines.append('')
+    words = '\n'.join(lines)
+    input_file = filename or mp3_file
+    # ffmpeg -i testing.mp4 -vf subtitles=transcript.srt mysubtitledmovie.mp4
+    # use ffmpeg bindings to add subtitles to video
+    # use python to call ffmpeg -i testing.mp4 -vf subtitles=transcript.srt mysubtitledmovie.mp4
+    
+    input_video = ffmpeg.input('testing.mp4')
+    subtitle = ffmpeg.filter('subtitles', 'transcript.srt')
+
+    output_video = ffmpeg.output(input_video, subtitle, 'subtitled.mp4', vcodec='libx264', video_filters='[v]subtitles=transcript.srt[v]')
+
+    ffmpeg.run(output_video)
+    # for spacy use advanced logic to extract and append to html_text using tables?
+
+    # get output_video as mp4
+    return result.get("segments"), words, "subtitled.mp4"
+ 
+gr.Interface(
+    title = 'Download Video From url and extract text from audio', 
+    fn=get_text_from_mp3_whisper, 
+    inputs=[
+        gr.Dropdown(["url", "file"]),
+        gr.inputs.Audio(type="filepath"),
+        gr.inputs.Textbox(),
+        gr.Dropdown(["translate", "transcribe"]),
+        gr.Dropdown(["Japanese", "English"])
+    ],
+    outputs=[
+        "json", "text", "mp4"
+    ],
+    live=True).launch()

requirements.txt

@@ -0,0 +1,4 @@
+openai-whisper
+youtube_dl
+ffmpeg-python
+gradio