app.py

import gradio as gr
import whisper
import os


class GradioInference():
    def __init__(self):
        self.sizes = ["base", "large"]
        self.langs = ["none"] + sorted(list(whisper.tokenizer.LANGUAGES.values()))
        self.current_size = "base"
        self.loaded_model = whisper.load_model(self.current_size)


    def __call__(self, file, lang, size, subs):
        path = file.name

        if lang == "none":
            lang = None

        if size != self.current_size:
            self.loaded_model = whisper.load_model(size)
            self.current_size = size
        results = self.loaded_model.transcribe(path, language=lang)

        if subs == "None":
            return results["text"]
        elif subs == ".srt":
            return self.srt(results["segments"])
        elif ".csv" == ".csv":
            return self.csv(results["segments"])

    def srt(self, segments):
        output = ""
        for i, segment in enumerate(segments):
            output += f"{i + 1}\n"
            output += f"{self.format_time(segment['start'])} --> {self.format_time(segment['end'])}\n"
            output += f"{segment['text']}\n\n"
        return output

    def csv(self, segments):
        output = ""
        for segment in segments:
            output += f"{segment['start']},{segment['end']},{segment['text']}\n"
        return output

    def format_time(self, time):
        hours = time // 3600
        minutes = (time - hours * 3600) // 60
        seconds = time - hours * 3600 - minutes * 60
        milliseconds = (time - int(time)) * 1000
        return f"{int(hours):02d}:{int(minutes):02d}:{int(seconds):02d},{int(milliseconds):03d}"


gio = GradioInference()
title = "Whisperer MP3"
description = "Transcription de fichiers MP3 en texte à l'aide du modèle Whisper d'OpenAI"

block = gr.Blocks()
with block:
    gr.HTML(
        """
            <div style="text-align: center; max-width: 500px; margin: 0 auto;">
              <div>
                <h1>Whisperer MP3</h1>
              </div>
              <p style="margin-bottom: 10px; font-size: 94%">
                Transcription de fichiers MP3 en texte à l'aide du modèle Whisper d'OpenAI
              </p>
            </div>
        """
    )
    with gr.Group():
        with gr.Box():
            with gr.Row().style(equal_height=True):
                sz = gr.Dropdown(label="Taille du modèle", choices=gio.sizes, value='base')
                lang = gr.Dropdown(label="Langue (facultatif)", choices=gio.langs, value="none")
            with gr.Row().style(equal_height=True):
                wt = gr.Radio(["None", ".srt", ".csv"], label="Avec horodatage ?")
            file = gr.File(label="Fichier MP3")
            text = gr.Textbox(label="Transcription", placeholder="Résultat de la transcription", lines=10)
            with gr.Row().style(equal_height=True):
                btn = gr.Button("Transcrire")       
btn.click(gio, inputs=[file, lang, sz, wt], outputs=[text])

block.launch()