app.py

# -*- coding: utf-8 -*-
"""Demo.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/14CzvGJOpNIGSyyH_fI5Q54ovm-TZCEHX
"""


import gradio as gr
from transformers import AutoProcessor
from transformers import WhisperForConditionalGeneration, WhisperProcessor
from pydub import AudioSegment
import torch
import os
import gradio as gr
import librosa
import numpy as np




# Model yükleme (CPU)
model_id = "SAadettin-BERber/whisper_small_model_atc_10"
adapter_id = "SAadettin-BERber/whisper-small_atc_10"

#merge yap
#model.load_adapter(adapter_id)

from transformers import pipeline

MODEL_OPTIONS = {
    "İnce Ayar Çekilmiş Whisper Small": "SAadettin-BERber/normalize_whisper-small_atc_shuffle_3",
    "İnce Ayar Çekilmiş Large Turbo": "SAadettin-BERber/normalize_whisper-large-v3-turbo_atc_shuffle_1",
    "Orijinal Whisper Small": "unsloth/whisper-small",
    "Orijinal Whisper Turbo": "unsloth/whisper-large-v3-turbo",
}

def transcribe_choice(audio_file, selected_model_name):
    model_id = MODEL_OPTIONS[selected_model_name]

    # Bilgilendirme
    #status_text = f"📦 '{selected_model_name}' modeli yükleniyor..."

    pipe = pipeline("automatic-speech-recognition", model=model_id, device=-1)

    audio = AudioSegment.from_file(audio_file)
    segment_length = 10_000  # 10 saniye
    results = []

    for i in range(0, len(audio), segment_length):
        segment = audio[i:i + segment_length]
        temp_path = f"temp_{i//1000}.wav"
        segment.export(temp_path, format="wav")

        output = pipe(temp_path)
        results.append(output["text"])

        os.remove(temp_path)

    transcription = " ".join(results)
    return transcription


with gr.Blocks() as demo:
    gr.Markdown("## ✨ Whisper Model Transcriber with Multiple Model Support")

    audio_input = gr.Audio(type="filepath", label="Ses Dosyası")
    model_choice = gr.Dropdown(choices=list(MODEL_OPTIONS.keys()), label="Model Seçin")
    #status_box = gr.Textbox(label="Durum", interactive=False)
    output_box = gr.Textbox(label="Transkripsiyon")

    transcribe_btn = gr.Button("Transkribe Et")

    transcribe_btn.click(
        fn=transcribe_choice,
        inputs=[audio_input, model_choice],
        outputs=[output_box]
    )

demo.launch()