Spaces:
Runtime error
Runtime error
File size: 3,057 Bytes
5c83790 bf76c51 99711d6 bf76c51 6a6cb15 4d91972 bf76c51 244dbca bf76c51 244dbca bf76c51 244dbca 99711d6 244dbca 99711d6 244dbca bf76c51 ceaf5e7 99711d6 fd34583 4d91972 5c83790 fd34583 5c83790 bf76c51 5c83790 99711d6 f22e219 99711d6 c8fce57 bf76c51 c8fce57 99711d6 244dbca 5c83790 244dbca 99711d6 244dbca bf76c51 244dbca c8fce57 bf76c51 244dbca c8fce57 244dbca bf76c51 5c83790 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 |
# MusicGen + Gradio + GPT Demo App (CPU-Optimized with MCP Server)
import gradio as gr
import os
import numpy as np
import torch
from transformers import AutoProcessor, MusicgenForConditionalGeneration
from openai import OpenAI
import scipy.io.wavfile
# Force CPU device (no GPU required)
device = torch.device("cpu")
# Load MusicGen model onto CPU
model_name = "facebook/musicgen-small"
model = MusicgenForConditionalGeneration.from_pretrained(model_name).to(device)
processor = AutoProcessor.from_pretrained(model_name)
# Initialize OpenAI client (set OPENAI_API_KEY in HF Spaces Secrets)
client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
# Refine user prompt via GPT
def refine_prompt(user_input):
completion = client.chat.completions.create(
model="gpt-4",
messages=[
{"role": "system", "content": "You are a music assistant. Make the user's input more descriptive for an AI music generator."},
{"role": "user", "content": user_input}
]
)
return completion.choices[0].message.content.strip()
# Generate music (shorter tokens for CPU speed)
def generate_music(prompt, max_new_tokens: int = 128):
inputs = processor(text=[prompt], return_tensors="pt").to(device)
audio_values = model.generate(**inputs, max_new_tokens=max_new_tokens)
sampling_rate = model.config.audio_encoder.sampling_rate
audio = audio_values[0].cpu().numpy()
# Normalize to float32 in -1.0 to 1.0 range for Gradio
audio = audio / np.max(np.abs(audio))
audio = audio.astype(np.float32)
# Prepare int16 version and ensure 1D for WAV
int_audio = (audio * 32767).astype(np.int16)
int_audio = np.squeeze(int_audio)
if int_audio.ndim > 1:
int_audio = int_audio[:, 0]
# Save as .wav file (in /tmp for Spaces)
scipy.io.wavfile.write("/tmp/output.wav", sampling_rate, int_audio)
return sampling_rate, audio
# Combined Gradio function
def main(user_input, max_new_tokens):
detailed_prompt = refine_prompt(user_input)
sampling_rate, audio = generate_music(detailed_prompt, max_new_tokens)
return detailed_prompt, (sampling_rate, audio), "/tmp/output.wav"
# Build Gradio UI
demo = gr.Blocks()
with demo:
gr.Markdown("""# 🎵 AI Music Generator
Enter a music idea or mood and get a short AI-generated track. (CPU mode)""")
user_input = gr.Textbox(label="Describe the mood or style of music")
max_tokens = gr.Slider(32, 256, value=128, step=32, label="Length (tokens) for CPU")
generate_btn = gr.Button("Generate Music")
refined_output = gr.Textbox(label="Enhanced Prompt by GPT")
audio_output = gr.Audio(label="Generated Audio", type="numpy")
download_wav = gr.File(label="Download .wav file")
generate_btn.click(
main,
inputs=[user_input, max_tokens],
outputs=[refined_output, audio_output, download_wav]
)
# Launch with Gradio MCP Server
from gradio.mcp_server import MCPServer
if __name__ == "__main__":
server = MCPServer(demo, host="0.0.0.0", port=7860)
server.run()
|