cocktailpeanut's picture
update
78d5a79
raw
history blame
7.24 kB
import gradio as gr
from musiclang_predict import MusicLangPredictor
from musiclang import Score
from midi2audio import FluidSynth
import os
import tempfile
def inner_loop(midi_file, chord_progression, tempo, temperature, nb_tokens, bar_range):
top_p = 0.98
seed = 0
# Initialize the MusicLangPredictor
ml = MusicLangPredictor('musiclang/musiclang-v2')
tempo_message = "" # Default message if no MIDI file is uploaded
time_signature = (4, 4)
if midi_file is not None and midi_file != "":
# Load the MIDI file and use it as the score prompt
filepath = midi_file
start_bar, end_bar = map(int, bar_range.split("-"))
score = Score.from_midi(filepath, chord_range=(start_bar, end_bar))
tempo = score.config['tempo'] # Use the tempo from the MIDI file and change input
time_signature = score.config['time_signature']
time_signature = (time_signature[1], time_signature[2])
tempo_message = f"Warning : real tempo of file is : {int(tempo)} BPM." # Update message based on MIDI file
else:
score = None # Default score is None if no MIDI file is uploaded
# Generate the score based on provided inputs and the uploaded MIDI file if available
if chord_progression.strip() == "" and score is None:
# Generate without specific chord progression or MIDI prompt
generated_score = ml.predict(
nb_tokens=int(nb_tokens),
temperature=float(temperature),
topp=top_p,
rng_seed=seed
)
elif score is not None and chord_progression.strip() == "":
# Generate using the uploaded MIDI file as a prompt
generated_score = ml.predict(
score=score, # Use the uploaded MIDI as the score prompt
nb_tokens=int(nb_tokens),
temperature=float(temperature),
topp=top_p,
rng_seed=seed
)
else:
# Generate with specific chord progression
generated_score = ml.predict_chords(
chord_progression,
score=score, # Use the uploaded MIDI as the score prompt
time_signature=time_signature,
temperature=temperature,
topp=top_p,
rng_seed=seed
)
chord_repr = generated_score.to_chord_repr()
# Save the generated score as a MIDI file
temp_midi_file = tempfile.NamedTemporaryFile(suffix=".mid", delete=False)
midi_path = temp_midi_file.name
generated_score.to_midi(midi_path, tempo=tempo, time_signature=time_signature)
# Convert MIDI to WAV then WAV to MP3 for playback
temp_wav_file = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
temp_mp3_file = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
wav_path = temp_wav_file.name
mp3_path = temp_mp3_file.name
FluidSynth("/usr/share/sounds/sf2/FluidR3_GM.sf2").midi_to_audio(midi_path, wav_path)
os.system(f'ffmpeg -i {wav_path} -acodec libmp3lame -y -loglevel quiet -stats {mp3_path}')
# Remove the temporary WAV file
os.remove(wav_path)
return mp3_path, midi_path, chord_repr, tempo_message
def musiclang(midi_file, chord_progression, tempo, temperature, nb_tokens, bar_range):
exception = None
mp3_path, midi_path, chord_repr, tempo_message = None, None, None, ""
try:
mp3_path, midi_path, chord_repr, tempo_message = inner_loop(midi_file, chord_progression, tempo, temperature, nb_tokens, bar_range)
except Exception as e:
exception = "Error : " + e.__class__.__name__ + " " + str(e)
# Return the MP3 path for Gradio to display and the MIDI file path for download
return mp3_path, midi_path, exception
with gr.Blocks() as demo:
# Introductory text
gr.Markdown("""
# Controllable Symbolic Music Generation with MusicLang Predict
[MusicLang Predict](https://github.com/musiclang/musiclang_predict) offers advanced controllability features and high-quality music generation by manipulating symbolic music.
You can for example use it to continue your composition with a specific chord progression.
""")
with gr.Row():
with gr.Column():
with gr.Row():
midi_file = gr.File(label="Prompt MIDI File (Optional)", type="filepath", file_types=[".mid", ".midi"],
elem_id='midi_file_input')
with gr.Column():
bar_range = gr.Textbox(label="Bar Range of input file (eg: 0-4 for first four bars)", placeholder="0-4",
value="0-4", elem_id='bar_range_input')
nb_tokens = gr.Number(label="Nb Tokens",
value=512, minimum=256, maximum=2048, step=256, elem_id='nb_tokens_input')
temperature = gr.Slider(
label="Temperature",
value=0.95,
visible=False,
minimum=0.1, maximum=1.0, step=0.1, elem_id='temperature_input')
tempo = gr.Slider(label="Tempo", value=120, minimum=60, maximum=240, step=1, elem_id='tempo_input')
with gr.Row():
chord_progression = gr.Textbox(
label="Chord Progression (Optional)",
placeholder="Am CM Dm7/F E7 Asus4", lines=2, value="", elem_id='chord_progression_input')
with gr.Row():
generate_btn = gr.Button("Generate", elem_id='generate_button')
with gr.Column():
info_message = gr.Textbox(label="Info Message", elem_id='info_message_output')
generated_music = gr.Audio(label="Preview generated Music", elem_id='generated_music_output')
generated_midi = gr.File(label="Download MIDI", elem_id='generated_midi_output')
generate_btn.click(
fn=musiclang,
inputs=[midi_file, chord_progression, tempo, temperature, nb_tokens, bar_range],
outputs=[generated_music, generated_midi, info_message]
)
with gr.Row():
with gr.Column():
gr.Markdown("## Examples")
gr.Examples(
examples=[[os.path.abspath("bach_847.mid"), "", 120, 0.95, 512, "0-4"],
[os.path.abspath("bach_847.mid"), "Cm C7/E Fm F#dim G7", 120, 0.95, 512, "0-4"],
[os.path.abspath("boney_m_ma_baker.mid"), "", 120, 0.95, 512, "0-4"],
[os.path.abspath("eminem_slim_shady.mid"), "Cm AbM BbM G7 Cm", 120, 0.95, 512, "0-4"],
[os.path.abspath("mozart_alla_turca.mid"), "", 120, 0.95, 512, "0-4"],
[os.path.abspath("mozart_alla_turca.mid"), "Am Em CM G7 E7 Am Am E7 Am", 120, 0.95, 512, "0-4"],
[os.path.abspath("daft_punk_around_the_world.mid"), "", 120, 0.95, 512, "0-4"],
],
inputs=[midi_file, chord_progression, tempo, temperature, nb_tokens, bar_range],
outputs=[generated_music, generated_midi, info_message],
fn=musiclang,
#cache_examples=True,
)
demo.launch()