app.py

import os
import sys
import gradio as gr
from inspiremusic.cli.inference import InspireMusicUnified, set_env_variables
import torchaudio
import datetime
import hashlib

def generate_filename():
	now = datetime.datetime.now()
	seconds_since_epoch = int(now.timestamp())
	# Convert seconds to string
	seconds_str = str(seconds_since_epoch)
	# Hash the string using SHA-256
	hash_object = hashlib.sha256(seconds_str.encode())
	hash_string = hash_object.hexdigest()
	return hash_string

def get_args(
		task, text="", audio=None, model_name="InspireMusic-Base",
		chorus="intro",
		output_sample_rate=48000, max_generate_audio_seconds=30.0, time_start = 0.0, time_end=30.0, trim=False):
	
	if output_sample_rate == 24000:
		fast = True
	else:
		fast = False
	# This function constructs the arguments required for InspireMusic
	args = {
		"task"                      : task,
		"text"                      : text,
		"audio_prompt"              : audio,
		"model_name"                : model_name,
		"chorus"                    : chorus,
		"fast"                      : fast,
		"fade_out"                  : True,
		"trim"                      : trim,
		"output_sample_rate"        : output_sample_rate,
		"min_generate_audio_seconds": 10.0,
		"max_generate_audio_seconds": max_generate_audio_seconds,
		"model_dir"                 : os.path.join("pretrained_models",
												   model_name),
		"result_dir"                : "exp/inspiremusic",
		"output_fn"                 : generate_filename(),
		"format"                    : "wav",
		"time_start" : time_start,
		"time_end": time_end,
		"fade_out_duration": 1.0,
	}

	if args["time_start"] is None:
		args["time_start"] = 0.0
	args["time_end"] = args["time_start"] + args["max_generate_audio_seconds"]

	print(args)
	return args


def music_generation(args):
	set_env_variables()
	model = InspireMusicUnified(
			model_name=args["model_name"],
			model_dir=args["model_dir"],
			min_generate_audio_seconds=args["min_generate_audio_seconds"],
			max_generate_audio_seconds=args["max_generate_audio_seconds"],
			sample_rate=24000,
			output_sample_rate=args["output_sample_rate"],
			load_jit=True,
			load_onnx=False,
			fast=args["fast"],
			result_dir=args["result_dir"])

	output_path = model.inference(
			task=args["task"],
			text=args["text"],
			audio_prompt=args["audio_prompt"],
			chorus=args["chorus"],
			time_start=args["time_start"],
			time_end=args["time_end"],
			output_fn=args["output_fn"],
			max_audio_prompt_length=args["max_audio_prompt_length"],
			fade_out_duration=args["fade_out_duration"],
			output_format=args["format"],
			fade_out_mode=args["fade_out"],
			trim=args["trim"])
	return output_path

def update_text():
    global text_input  # Declare as global to modify the outer scope variable
    text_input = "New value set by button click"
    return text_input

default_prompts = [
    "Experience soothing and sensual instrumental jazz with a touch of Bossa Nova, perfect for a relaxing restaurant or spa ambiance.",
	"Compose an uplifting R&B song.",
	"Create an emotional, introspective folk song with acoustic guitar and soft vocals."
]

def cut_audio(audio_file, cut_seconds=5):
	audio, sr = torchaudio.load(audio_file)
	num_samples = cut_seconds * sr
	cutted_audio = audio[:, :num_samples]
	output_path = os.path.join(os.getcwd(), "audio_prompt_" + generate_filename() + ".wav")
	torchaudio.save(output_path, cutted_audio, sr)
	return output_path

def run_text2music(text, model_name, chorus,
					 output_sample_rate, max_generate_audio_seconds):
	args = get_args(
			task='continuation', text=text, audio=None,
			model_name=model_name, chorus=chorus,
			output_sample_rate=output_sample_rate,
			max_generate_audio_seconds=max_generate_audio_seconds)
	return music_generation(args)

def run_continuation(text, audio, model_name, chorus,
					 output_sample_rate, max_generate_audio_seconds):
	args = get_args(
			task='text-to-music', text=text, audio=cut_audio(audio, cut_seconds=5),
			model_name=model_name, chorus=chorus,
			output_sample_rate=output_sample_rate,
			max_generate_audio_seconds=max_generate_audio_seconds)
	return music_generation(args)

with gr.Blocks(theme=gr.themes.Soft()) as demo:
	gr.Markdown("""
    # InspireMusic
    - Support text-to-music, music continuation, audio super-resolution, audio reconstruction tasks with high audio quality, with available sampling rates of 24kHz, 48kHz. 
    - Support long audio generation in multiple output audio formats, i.e., wav, flac, mp3, m4a.
    - Open-source [InspireMusic-Base](https://modelscope.cn/models/iic/InspireMusic/summary), [InspireMusic-Base-24kHz](https://modelscope.cn/models/iic/InspireMusic-Base-24kHz/summary), [InspireMusic-1.5B](https://modelscope.cn/models/iic/InspireMusic-1.5B/summary), [InspireMusic-1.5B-24kHz](https://modelscope.cn/models/iic/InspireMusic-1.5B-24kHz/summary), [InspireMusic-1.5B-Long](https://modelscope.cn/models/iic/InspireMusic-1.5B-Long/summary) models for music generation.
    - Currently only support English text prompts.
    """)

	with gr.Row(equal_height=True):
		model_name = gr.Dropdown(["InspireMusic-1.5B-Long", "InspireMusic-1.5B", "InspireMusic-1.5B-24kHz", "InspireMusic-Base", "InspireMusic-Base-24kHz"], label="Select Model Name", value="InspireMusic-Base")
		chorus = gr.Dropdown(["intro", "verse", "chorus", "outro"],
							 label="Chorus Mode", value="intro")
		output_sample_rate = gr.Dropdown([48000, 24000],
										 label="Output Audio Sample Rate (Hz)",
										 value=48000)
		max_generate_audio_seconds = gr.Slider(10, 120,
											   label="Generate Audio Length (s)",
											   value=30)
		# with gr.Column():
		# 	fast = gr.Checkbox(label="Fast Inference", value=False)
		# 	fade_out = gr.Checkbox(label="Apply Fade Out Effect", value=True)

	with gr.Row(equal_height=True):
		# Textbox for custom input
		text_input = gr.Textbox(label="Input Text (For Text-to-Music Task)", value="Experience soothing and sensual instrumental jazz with a touch of Bossa Nova, perfect for a relaxing restaurant or spa ambiance.")

		audio_input = gr.Audio(label="Input Audio Prompt (For Music Continuation Task)",
							   type="filepath")
	music_output = gr.Audio(label="Generated Music", type="filepath")

	with gr.Row():
		button = gr.Button("Text to Music")
		button.click(run_text2music,
						  inputs=[text_input, model_name,
								  chorus,
								  output_sample_rate,
								  max_generate_audio_seconds],
						  outputs=music_output)

		generate_button = gr.Button("Music Continuation")
		generate_button.click(run_continuation,
						  inputs=[text_input, audio_input, model_name,
								  chorus,
								  output_sample_rate,
								  max_generate_audio_seconds],
						  outputs=music_output)

	with gr.Column():
		default_prompt_buttons = []
		for prompt in default_prompts:
			button = gr.Button(value=prompt)
			button.click(run_text2music,
						 inputs=[text_input, model_name,
								 chorus,
								 output_sample_rate,
								 max_generate_audio_seconds],
						 outputs=music_output)
			default_prompt_buttons.append(button)
demo.launch()