app.py

import os
import torch
import gradio as gr
from transformers import pipeline
from huggingface_hub import InferenceClient

# ----------------------
# AUDIO-TO-TEXT SETUP
# ----------------------
device = 0 if torch.cuda.is_available() else "cpu"
AUDIO_MODEL_NAME = "distil-whisper/distil-large-v3"
BATCH_SIZE = 8

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

def transcribe(audio_input):
    """Convert audio to text using Whisper."""
    if audio_input is None:
        raise gr.Error("No audio file submitted!")

    output = pipe(
        audio_input,
        batch_size=BATCH_SIZE,
        generate_kwargs={"task": "transcribe"},
        return_timestamps=True
    )
    return output["text"]

# ----------------------
# TEXT ORGANIZATION SETUP
# ----------------------
TEXT_MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"

# Ensure HF_TOKEN is loaded as a Space secret
hf_token = os.getenv("HF_TOKEN")
if not hf_token:
    raise ValueError("HF_TOKEN not found! Add it as a secret in your Space settings.")

# Force client to use the HF inference API
client = InferenceClient(token=hf_token, base_url="https://api-inference.huggingface.co")

def build_messages(meeting_transcript) -> list:
    system_input = "You are an assistant that organizes meeting minutes."
    user_input = f"""
    Take this raw meeting transcript and return an organized, sectioned version.
    You may include a summary at the top.

    Transcript:
    {meeting_transcript}
    """
    return [
        {"role": "system", "content": system_input},
        {"role": "user", "content": user_input},
    ]

def organize_text(meeting_transcript):
    messages = build_messages(meeting_transcript)
    response = client.chat_completion(
        messages, model=TEXT_MODEL_NAME, max_tokens=300, seed=42
    )
    return response.choices[0].message.content

# ----------------------
# COMBINED TOOL
# ----------------------
def meeting_transcript_tool(audio_input):
    meeting_text = transcribe(audio_input)
    organized_text = organize_text(meeting_text)
    return organized_text

# ----------------------
# GRADIO INTERFACE
# ----------------------
demo = gr.Interface(
    fn=meeting_transcript_tool,
    inputs=gr.Audio(type="filepath"),
    outputs=gr.Textbox(show_copy_button=True, label="Organized Transcript"),
    title="🪶 Meeting Transcription Tool",
    description="Upload or record an audio file. This app transcribes it using Whisper and organizes the text using Phi-3",
)

demo.launch()