app.py

from transformers import pipeline
import gradio as gr
import moviepy.editor as mp
from pytube import YouTube
import math
import youtube_dl
import validators

pipe = pipeline(model="errno98/whisper-small-hi")

segment_len = 30

def download_soundcloud(url):
    
    options = {
      'format': 'bestaudio/best',
      'extractaudio' : True,  # only keep the audio
      'audioformat' : "mp3",  # convert to mp3 
      'outtmpl': '%(id)s',    # name the file the ID of the video
      'noplaylist' : True,    # only download single song, not playlist
    }

    with youtube_dl.YoutubeDL() as ydl:
        ydl.download([url])
        
    # download metadata
    ydl = youtube_dl.YoutubeDL(options)

    r = None
    url = url
    with ydl:
        # don't download, much faster 
        r = ydl.extract_info(url, download=True)  

    global video_clip
    global audio_wav
    global audio_len
    
    #my_clip = mp.VideoFileClip(r)
    #my_clip.audio.write_audiofile("audio.wav")
    #audio_wav = mp.AudioFileClip("audio.wav")
    #audio_len = audio_wav.duration
    
    return r['title']

def load_vid(url):
    video = (
        YouTube(url)
        .streams.filter(progressive=True, file_extension="mp4")
        .first()
        .download()
    )
    
    global video_clip
    global audio_wav
    global audio_len
    
    video_clip = mp.VideoFileClip(video)
    video_clip.audio.write_audiofile("audio.wav")
    audio_wav = mp.AudioFileClip("audio.wav")
    audio_len = audio_wav.duration
    
    return video

def validate_link(url):

    try:
        yt = YouTube(url)  
    except Exception:
        return True

    video_length = yt.length
    if video_length > 600:
        print("Video length is too long (longer than 10 minutes)")
        return False
    else:
        return True

def clippify(index, seg_total):
    audio_file = "audio_out"+str(index)+".wav"
    audio_clipped_obj = mp.AudioFileClip.copy(audio_wav)
    
    if (index > 0):
        audio_clipped_obj = mp.AudioFileClip.cutout(audio_clipped_obj, 0, segment_len * (index))
    if (index < seg_total - 1):
        audio_clipped_obj = mp.AudioFileClip.cutout(audio_clipped_obj, segment_len * (index + 1), audio_length)
    
    mp.AudioFileClip.write_audiofile(audio_clipped_obj, audio_file)
    
    return audio_file

def transcribe(url):

    if url[0:8] == "https://" or url[0:8] == "http://":
        url = url[8::]
    
    text = ""

    if not validators.url(url):
        if not validate_link(url):
            return "Not a YouTube video"
        else:
            load_vid(url)
    else: 
        return "Incorrect URL structure"
    
    
    segment_count = math.ceil(audio_len / segment_len) 

    if segment_count <= 0:
        return "Invalid segment length"
    else:
        for x in range(segment_count):
            audio = clippify(x, segment_count)
            seg_text = pipe(audio, batch_size=512, truncation=True)["text"]
            print("Segtext: ")
            print(seg_text)
            text = text + seg_text
            
    return text

iface = gr.Interface(
    fn=transcribe, 
    inputs=gr.Textbox(label = "Input the URL of a YouTube video:"), 
    outputs="text",
    title="Whisper Small SE",
    description="Video Swedish Transcription",
)

iface.launch()