app.py

import gradio as gr
import torch
from faster_whisper import WhisperModel
import pandas as pd

# Model size selection
model_size = "large-v2"

# Get device
device = "cuda:0" if torch.cuda.is_available() else "cpu"

# Initialize model based on device
if device == "cuda:0":
    # Run on GPU with FP16
    model_whisper = WhisperModel(model_size, device="cuda", compute_type="float16")
else:
    # Run on CPU with INT8
    model_whisper = WhisperModel(model_size, device="cpu", compute_type="int8")

# Function to get filename from file object
def get_filename(file_obj):
    return file_obj.name.split("/")[-1]

# Function to transcribe audio to text
def audio_to_transcript(file_obj):
    try:
        filename = get_filename(file_obj)
        segments, _ = model_whisper.transcribe(file_obj.name, beam_size=5, vad_filter=True)
    except:
        filename = file_obj.split("/")[-1]
        segments, _ = model_whisper.transcribe(file_obj, beam_size=5, vad_filter=True)

    # Initialize lists to store transcription data
    start_segments, end_segments, text_segments = list(), list(), list()
    
    # Process each segment for start time, end time, and text
    for segment in segments:
        start, end, text = segment.start, segment.end, segment.text
        start_segments.append(start)
        end_segments.append(end)
        text_segments.append(text)

    # Save transcript to CSV
    df = pd.DataFrame()
    df["start"] = start_segments
    df["end"] = end_segments
    df["text"] = text_segments

    # Define output CSV file
    csv_file = filename.split(".")[0] + ".csv"
    df.to_csv(csv_file, encoding="utf-8", index=False)
    path_to_csv = gr.File.update(value=csv_file, visible=True)

    return filename, path_to_csv, df

## Gradio Interface Setup
headers = ["start", "end", "text"]

iface = gr.Interface(
    fn=audio_to_transcript,
    inputs=gr.File(label="Upload an Audio File", type="filepath"),
    outputs=[
        gr.Textbox(label="Audio file name"),
        gr.File(label="Transcript CSV file"),
        gr.DataFrame(label="Transcript", headers=headers),
    ],
    allow_flagging="never",
    title="Audio to Transcript",
    description="Upload an audio file, and this tool will return a transcript with time-stamped segments.",
    theme="compact",  # Enhanced UI theme for simplicity
)

iface.launch(debug=True)