app.py

import whisper
from whisper.transcribe import LANGUAGES, TO_LANGUAGE_CODE
from whisper.utils import get_writer
import torch
import spaces
import gradio as gr
import pathlib
import random
from datetime import datetime

APP_DIR = pathlib.Path(__file__).parent.absolute()

LOCAL_DIR = APP_DIR / "files"
LOCAL_DIR.mkdir(exist_ok=True)
SAVE_DIR = LOCAL_DIR / "transcripts"
SAVE_DIR.mkdir(exist_ok=True)
LANGS = [lang.capitalize() for lang in list(LANGUAGES.values())]
# DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
DEVICE = 'cuda'
loaded_model = whisper.load_model("medium", "cpu")
current_size = "None"

def generate_random_filename():
    now = datetime.now()
    timestamp = now.strftime("%H_%M_%S_%d_%m_%Y")
    random_suffix = ''.join(random.choices('abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789', k=8))
    filename = f"transcript_{timestamp}_{random_suffix}"
    return filename

@spaces.GPU(enable_queue=True)
def get_transcript(audio_path, task_selection:str, language:str, output_format:str, max_line_width=0, max_line_count=0, max_words_per_line=0):
    writer = get_writer(output_format.lower(), SAVE_DIR)
    writer_args = {
        "max_line_width": max_line_width if max_line_width > 0 else None,
        "max_line_count": max_line_count if max_line_count > 0 else None,
        "max_words_per_line": max_words_per_line if max_words_per_line > 0 else None

    }

    options = dict(task=task_selection.lower(), best_of=5, language=TO_LANGUAGE_CODE[language.lower()])
    loaded_model.to(DEVICE)
    results = loaded_model.transcribe(audio_path, verbose=True, word_timestamps=True, **options)
    filename = generate_random_filename()
    writer(results, filename, **writer_args)

    return str(SAVE_DIR / f"{filename}.{output_format.lower()}")

title="""
<div style="text-align: center; max-width: 500px; margin: 0 auto;">
        <div
        style="
            display: inline-flex;
            align-items: center;
            gap: 0.8rem;
            font-size: 1.75rem;
            margin-bottom: 10px;
        "
        >
        <h1 style="font-weight: 600; margin-bottom: 7px;">
            Auto Transcriber 🔊
        </h1>
        </div>
       
    </div>
"""

with gr.Blocks() as monapp:
    with gr.Column():
        gr.HTML(title)
        with gr.Row():
            with gr.Column():
                audio_input = gr.Audio(label="Audiofile to transcribe", sources=["upload"], type="filepath")
                with gr.Accordion("Transcribe options", open=True):
                    task_selection = gr.Radio(["Transcribe", "Translate"], value="Transcribe", label="Select a Task")
                    language = gr.Dropdown(choices=LANGS, value="English", label="Language spoken in the audio")
                    output_format = gr.Radio(["TXT", "VTT", "SRT", "TSV", "JSON"], value="TXT", label="Format of the output file")
                    with gr.Column():
                        gr.HTML("<p>keep at 0 to <strong>don't use</strong></p>\n<p>max_words_per_line has no effect with max_line_width activated\nWord-level timestamps on translations may not be reliable.</p>")
                        max_line_width = gr.Number(label="Maximum number of characters in a line before breaking the line", minimum=0, precision=0, value=0, step=1)
                        max_line_count = gr.Number(label="Maximum number of lines in a segment", minimum=0, precision=0, value=0, step=1)
                        max_words_per_line = gr.Number(label="Maximum number of words in a segment", minimum=0, precision=0, value=0, step=1)
                submit_btn = gr.Button("Transcribe")
            
            with gr.Column():
                transcript = gr.File(height=50)

    submit_btn.click(fn=get_transcript, inputs=[audio_input, task_selection, language, output_format, max_line_width, max_line_count, max_words_per_line], outputs=[transcript])

monapp.launch(debug=True, show_error=True)