app.py

import gradio as gr
import torch
from transformers import pipeline
import librosa
import os

# Config
MODEL_NAME = "openai/whisper-small"
BATCH_SIZE = 8
YT_LENGTH_LIMIT_S = 3600  # 1 hour
device = 0 if torch.cuda.is_available() else "cpu"

# Load model pipeline
pipe = pipeline(
    task="automatic-speech-recognition",
    model=MODEL_NAME,
    chunk_length_s=30,
    device=device,
)

# Duration check
def is_too_long(audio_path, limit=YT_LENGTH_LIMIT_S):
    duration = librosa.get_duration(path=audio_path)
    return duration > limit

# Transcription logic
def transcribe(audio_path, task="transcribe"):
    if audio_path is None or not os.path.exists(audio_path):
        raise gr.Error("❌ Invalid file path or missing audio.")

    if is_too_long(audio_path):
        raise gr.Error("⚠️ Audio exceeds 1-hour limit. Please upload a shorter file.")

    result = pipe(audio_path, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
    return result["text"]

# Wrapper with error handling
def handle_audio(audio_path, task):
    try:
        return transcribe(audio_path, task)
    except Exception as e:
        return f"❌ Error: {str(e)}"

# Gradio App
with gr.Blocks() as demo:
    gr.Markdown("# 🎙️ Whisper Speech-to-Text App\nPowered by 🤗 Transformers and Gradio.\n\nUpload an audio file or record with your mic.")

    with gr.Tabs():
        # Tab 1: Upload
        with gr.Tab("📁 Upload Audio"):
            with gr.Row():
                audio_upload = gr.Audio(type="filepath", label="Upload Audio File")
                task_option_upload = gr.Radio(["transcribe", "translate"], value="transcribe", label="Choose Task")
            with gr.Row():
                transcribe_btn_upload = gr.Button("Transcribe")
                output_upload = gr.Textbox(label="📝 Transcription", lines=8)
            transcribe_btn_upload.click(handle_audio, inputs=[audio_upload, task_option_upload], outputs=output_upload)

            gr.Examples(
                examples=[
                    "sample_audio/sample1.wav",
                    "sample_audio/sample2.mp3"
                ],
                inputs=[audio_upload],
                label="Example Audio Files"
            )

        # # Tab 2: Record
        # with gr.Tab("🎤 Record Audio"):
        #     with gr.Row():
        #         audio_record = gr.Audio(source="microphone", type="filepath", label="Record with Microphone")
        #         task_option_record = gr.Radio(["transcribe", "translate"], value="transcribe", label="Choose Task")
        #     with gr.Row():
        #         transcribe_btn_record = gr.Button("Transcribe")
        #         output_record = gr.Textbox(label="📝 Transcription", lines=8)
        #     transcribe_btn_record.click(handle_audio, inputs=[audio_record, task_option_record], outputs=output_record)

demo.launch()