This repository contains a demo for a MIDI composer model. The setup instructions below will guide you through the process of setting up the environment and running the demo.
-
Clone the Repository: First, clone the repository from Hugging Face Spaces:
git clone https://huggingface.co/spaces/skytnt/midi-composer
-
Navigate to the Repository Directory: Change to the cloned repository directory:
cd midi-composer -
Install Required Packages: Install the necessary system packages and Python dependencies:
sudo apt-get update && sudo apt-get install fluidsynth pip install -r requirements.txt pip install spaces
Once the environment is set up, you can run the demo using the following command:
python app.py --max-gen 4096 --shareThis command will start the application with a maximum generation limit of 4096 and share the application publicly.
from gradio_client import Client, handle_file
def generate_and_render_audio(model_name, instruments, drum_kit, input_midi_path=None, bpm=0, time_sig="auto", key_sig="auto", midi_events=128, gen_events=2048, temp=1.0, top_p=0.95, top_k=20, should_render_audio=True):
"""
生成 MIDI 并渲染音频的函数。
参数:
model_name (str): 模型名称。
instruments (list): 乐器列表。
drum_kit (str): 鼓组名称。
input_midi_path (str): 输入 MIDI 文件路径（可选）。
bpm (int): 每分钟节拍数（0 表示自动）。
time_sig (str): 时间签名（如 "4/4"）。
key_sig (str): 调号（如 "C"）。
midi_events (int): 使用前 n 个 MIDI 事件作为提示。
gen_events (int): 生成的最大 MIDI 事件数。
temp (float): 温度参数。
top_p (float): top-p 采样参数。
top_k (int): top-k 采样参数。
should_render_audio (bool): 是否渲染音频。
返回:
tuple: 包含生成的音频文件路径的元组。
"""
# 初始化客户端
client = Client("http://127.0.0.1:7860")
#client = Client("https://36a57c505482588881.gradio.live")
# 调用 /run API 生成 MIDI
run_result = client.predict(
model_name=model_name,
continuation_select="all",
instruments=instruments,
drum_kit=drum_kit,
bpm=bpm,
time_sig=time_sig,
key_sig=key_sig,
mid=handle_file(input_midi_path) if input_midi_path else None,
midi_events=midi_events,
reduce_cc_st=True,
remap_track_channel=True,
add_default_instr=True,
remove_empty_channels=False,
seed=0,
seed_rand=True,
gen_events=gen_events,
temp=temp,
top_p=top_p,
top_k=top_k,
allow_cc=True,
api_name="/run"
)
# 提取生成的 MIDI 序列
mid_seq = run_result[1]
# 调用 /render_audio API 渲染音频
render_result = client.predict(
model_name=model_name,
should_render_audio=should_render_audio,
api_name="/render_audio"
)
# 返回渲染的音频文件路径
return render_result
# 示例参数
model_name = "generic pretrain model (tv2o-medium) by skytnt"
instruments = ["Acoustic Grand", "Electric Piano 1"]
drum_kit = "Standard"
#input_midi_path = "path/to/your/input.mid" # 可选
bpm = 120
time_sig = "4/4"
key_sig = "C"
midi_events = 128
gen_events = 2048
temp = 1.0
top_p = 0.95
top_k = 20
should_render_audio = True
# 调用函数生成并渲染音频
audio_files = generate_and_render_audio(
model_name=model_name,
instruments=instruments,
drum_kit=drum_kit,
#input_midi_path=input_midi_path,
bpm=bpm,
time_sig=time_sig,
key_sig=key_sig,
midi_events=midi_events,
gen_events=gen_events,
temp=temp,
top_p=top_p,
top_k=top_k,
should_render_audio=should_render_audio
)
# 输出音频文件路径
print("Generated audio files:", audio_files)
import os
import shutil
def copy_and_rename_audio_files(audio_files, target_folder, prefix="output"):
"""
将音频文件从临时文件夹拷贝到目标文件夹并重命名。
参数:
audio_files (tuple): 包含音频文件路径的元组。
target_folder (str): 目标文件夹路径。
prefix (str): 重命名文件的前缀，默认为 "output"。
返回:
list: 包含新文件路径的列表。
"""
# 确保目标文件夹存在
os.makedirs(target_folder, exist_ok=True)
# 存储新文件路径
new_file_paths = []
# 遍历音频文件
for i, audio_file in enumerate(audio_files):
if audio_file: # 确保文件路径不为空
# 生成新文件名
new_file_name = f"{prefix}_{i + 1}.mp3"
new_file_path = os.path.join(target_folder, new_file_name)
# 拷贝文件
shutil.copy(audio_file, new_file_path)
new_file_paths.append(new_file_path)
return new_file_paths
copy_and_rename_audio_files(audio_files, target_folder = "demo0", prefix="output")- MIDI File Render
- https://huggingface.co/spaces/asigalov61/Advanced-MIDI-Renderer
- v1.3: MIDITokenizerV2 and new MidiVisualizer
- v1.2 : Optimise the tokenizer and dataset. The dataset was filtered by MIDITokenizer.check_quality. Using the higher quality dataset to train the model, the performance of the model is significantly improved.
projectlosangeles/Los-Angeles-MIDI-Dataset
- install pytorch(recommend pytorch>=2.0)
- install fluidsynth>=2.0.0
pip install -r requirements.txt
python app.py
python train.py
@misc{skytnt2024midimodel,
author = {SkyTNT},
title = {Midi Model: Midi event transformer for symbolic music generation},
year = {2024},
howpublished = {\url{https://github.com/SkyTNT/midi-model}},
}