app.py

import os
import json
import gradio as gr
import requests
from dotenv import load_dotenv
from datetime import datetime
from pathlib import Path
from basic_pitch.inference import predict_and_save
from basic_pitch import ICASSP_2022_MODEL_PATH
from music21 import converter
import base64

# === 1. Environment Configuration & OpenAI Client ===
load_dotenv()
OPENAI_API_KEY    = os.getenv("OPENAI_API_KEY")
MUSICGEN_API_URL  = os.getenv("MUSICGEN_API_URL")
VEROVIO_API_URL   = os.getenv("VEROVIO_API_URL")
assert OPENAI_API_KEY, "❌ Please set OPENAI_API_KEY in your .env file"

# Use OpenAI v1 client
from openai import OpenAI
openai_client = OpenAI(api_key=OPENAI_API_KEY)

# Create output directory if it doesn't exist
Path("output").mkdir(exist_ok=True)

# === 2. Tool Functions ===

def generate_music_from_hum(melody_file: str, prompt: str) -> str:
    """
    Call an external MusicGen API to generate a music WAV file
    based on a user’s humming audio and a style prompt.
    """
    if not MUSICGEN_API_URL:
        return "❌ MUSICGEN_API_URL is not configured"
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    output_wav = f"output/generated_{timestamp}.wav"
    try:
        with open(melody_file, "rb") as f:
            files = {"melody": ("hum.wav", f, "audio/wav")}
            data = {"text": prompt}
            response = requests.post(MUSICGEN_API_URL, files=files, data=data, timeout=180)
        if response.status_code != 200:
            return f"❌ MusicGen API error {response.status_code}: {response.text}"
        with open(output_wav, "wb") as out:
            out.write(response.content)
        return output_wav
    except Exception as e:
        return f"❌ Music generation failed: {e}"

def wav_to_musicxml(wav_path: str, timestamp: str=None) -> str:
    """
    Convert a WAV audio file to MusicXML using basic-pitch for pitch detection.
    """
    ts = timestamp or datetime.now().strftime("%Y%m%d_%H%M%S")
    # Remove any old MIDI files
    for midi_file in Path("output").glob("*_basic_pitch.mid"):
        midi_file.unlink()
    # Generate MIDI from the WAV
    predict_and_save(
        audio_path_list=[wav_path],
        output_directory="output",
        save_midi=True,
        sonify_midi=False,
        save_model_outputs=False,
        save_notes=False,
        model_or_model_path=ICASSP_2022_MODEL_PATH
    )
    midi_files = list(Path("output").glob("*.mid"))
    if not midi_files:
        return "❌ Failed to generate MIDI file"
    score = converter.parse(str(midi_files[0]))
    xml_path = f"output/generated_{ts}.musicxml"
    score.write("musicxml", fp=xml_path)
    return xml_path

def render_musicxml_via_verovio_api(musicxml_path: str) -> str:
    """
    Render a MusicXML file to an SVG preview using the Verovio API.
    Returns HTML containing the embedded SVG.
    """
    if not VEROVIO_API_URL:
        return "❌ VEROVIO_API_URL is not configured"
    try:
        with open(musicxml_path, "rb") as f:
            response = requests.post(VEROVIO_API_URL, files={"file": f}, timeout=120)
        if response.status_code != 200:
            return f"❌ Verovio API error {response.status_code}: {response.text}"
        svg = response.json().get("svg", "")
        b64_svg = base64.b64encode(svg.encode("utf-8")).decode("utf-8")
        return (
            '<div style="background:white;padding:10px;border-radius:8px;">'
            f'<img src="data:image/svg+xml;base64,{b64_svg}" style="width:100%;" />'
            '</div>'
        )
    except Exception as e:
        return f"❌ SVG rendering failed: {e}"

def generate_score_from_audio(wav_file: str) -> str:
    """
    Extract a MusicXML score from a generated music WAV file.
    """
    try:
        return wav_to_musicxml(wav_file)
    except Exception as e:
        return f"❌ Score extraction failed: {e}"

# Map of tool names to functions
TOOL_MAP = {
    "generate_music_from_hum": generate_music_from_hum,
    "wav_to_musicxml": wav_to_musicxml,
    "render_musicxml_via_verovio_api": render_musicxml_via_verovio_api,
    "generate_score_from_audio": generate_score_from_audio,
}

# === 3. GPT Tool Selection ===

def gpt_decide_tool(message: str, audio_path: str) -> dict:
    system_prompt = """
You are an AI music assistant. The user uploads an audio file and provides a request.
Choose the most appropriate tool from the list below and respond with strict JSON:

- generate_music_from_hum(melody_file, prompt)
- wav_to_musicxml(wav_file)
- render_musicxml_via_verovio_api(musicxml_file)
- generate_score_from_audio(wav_file)

JSON format:
{
  "tool_name": "...",
  "args": { ... },
  "explanation": "Reasoning explanation"
}
"""
    user_prompt = f"User request: {message}\nAudio file path: {audio_path}"
    response = openai_client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "system", "content": system_prompt},
            {"role": "user",   "content": user_prompt}
        ],
        temperature=0.2
    )
    text = response.choices[0].message.content
    try:
        return json.loads(text)
    except Exception:
        return {"error": f"Failed to parse JSON from GPT response:\n{text}"}

# === 4. Main Logic & Dynamic Output Display ===

def handle_request(audio_file, user_prompt):
    # Input validation
    if not audio_file or not user_prompt:
        return (
            "❗ Please upload an audio file and enter a request",
            "", "",
            gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
        )
    plan = gpt_decide_tool(user_prompt, audio_file)
    if "error" in plan:
        return (plan["error"], "", "") + (gr.update(visible=False),)*3

    tool_name  = plan["tool_name"]
    args       = plan.get("args", {})
    explanation= plan.get("explanation", "")
    log        = f"🧠 GPT chose: {tool_name}\n📦 Args: {json.dumps(args, ensure_ascii=False, indent=2)}"

    fn = TOOL_MAP.get(tool_name)
    if not fn:
        return (f"❌ Unknown tool: {tool_name}", explanation, log) + (gr.update(visible=False),)*3

    output = fn(**args)

    # Determine output type and update components accordingly
    if isinstance(output, str) and output.endswith(".wav") and os.path.isfile(output):
        return (
            "✅ Success", explanation, log,
            gr.update(value=output, visible=True),  # Audio
            gr.update(visible=False),               # SVG
            gr.update(visible=False)                # Text
        )
    if isinstance(output, str) and output.endswith(".musicxml") and os.path.isfile(output):
        # Automatically render MusicXML to SVG
        svg_html = render_musicxml_via_verovio_api(output)
        return (
            "✅ Success", explanation, log,
            gr.update(visible=False),
            gr.update(value=svg_html, visible=True),
            gr.update(visible=False)
        )
    if isinstance(output, str) and output.strip().startswith("<div"):
        # Already HTML SVG
        return (
            "✅ Success", explanation, log,
            gr.update(visible=False),
            gr.update(value=output, visible=True),
            gr.update(visible=False)
        )
    # Otherwise treat as plain text
    return (
        "✅ Success", explanation, log,
        gr.update(visible=False),
        gr.update(visible=False),
        gr.update(value=str(output), visible=True)
    )

# === 5. Gradio Interface ===

with gr.Blocks(title="🎶 Vibe Jamming – Your Music Assistant") as demo:
    gr.Markdown("## 🎵 Vibe Jamming – Your Music Assistant")

    with gr.Row():
        audio_input = gr.Audio(label="Upload Audio (.wav)", type="filepath")
        text_input  = gr.Textbox(label="Your Request", placeholder="e.g., Generate jazz music from my humming")

    run_button = gr.Button("🚀 Run")
    status_box  = gr.Textbox(label="Status")
    explanation_box = gr.Textbox(label="Explanation")
    log_box     = gr.Textbox(label="Tool Log", lines=6)

    audio_output = gr.Audio(label="🎧 Audio Output", visible=False, type="filepath")
    svg_output   = gr.HTML(label="🖼️ Score Preview (SVG)", visible=False)
    text_output  = gr.Textbox(label="📄 Text Output", visible=False, lines=4)

    run_button.click(
        fn=handle_request,
        inputs=[audio_input, text_input],
        outputs=[status_box, explanation_box, log_box, audio_output, svg_output, text_output]
    )

if __name__ == "__main__":
    demo.launch()