app.py

#====================================================================
# https://huggingface.co/spaces/projectlosangeles/Orpheus-MIDI-Search
#====================================================================

"""
Search for similar MIDIs with Orpheus embeddings
"""

#====================================================================

print('=' * 70)
print("Orpheus MIDI Search Gradio App")
print('=' * 70)
print("Loading modules...")

import os

os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

import time as reqtime
import datetime
from pytz import timezone

import numpy as np

from sentence_transformers import SentenceTransformer, util

import matplotlib.pyplot as plt

import gradio as gr

from huggingface_hub import hf_hub_download

import TMIDIX

from midi_to_colab_audio import midi_to_colab_audio

#====================================================================

PDT = timezone('US/Pacific')

MODEL_CHECKPOINT = 'all-mpnet-base-v2'
EMB_DATASETS_REPO = 'projectlosangeles/Orpheus-MIDI-Search'

MI_EMB_DATASET_FILES = ['168082_Orpheus_Song_Artist_Cap_Score_Sim_Dataset_CC_BY_NC_SA.pickle',
                        '168082_orpheus_song_artist_corpus_emb_all_mpnet_base_v2.npy'
                       ]
SP_EMB_DATASET_FILES = ['164598_Orpheus_Piano_Song_Artist_Cap_Score_Sim_Dataset_CC_BY_NC_SA.pickle',
                        '164598_orpheus_piano_song_artist_corpus_emb_all_mpnet_base_v2.npy'
                       ]

SOUNDFONT_PATH = 'SGM-v2.01-YamahaGrand-Guit-Bass-v2.7.sf2'

#====================================================================

dataset = 'Multi-Instrumental' # You can switch it to Piano one if you want

#====================================================================

print('=' * 70)
print("Done loading modules!")
print('=' * 70)

#====================================================================

print('Loading Sentence Transformer model...')
print('=' * 70)
model = SentenceTransformer(MODEL_CHECKPOINT)
print('=' * 70)
print('Done!')
print('=' * 70)

#====================================================================

# -----------------------------
# RENDER FUNCTION
# -----------------------------
def render_midi_output(final_composition, input_title=''):

    midi_score = save_midi(final_composition,
                           input_title=input_title
                          )
    
    midi_plot = TMIDIX.plot_ms_SONG(midi_score,
                                    plot_title=input_title,
                                    return_plt=True
                                   )
    
    midi_audio = midi_to_colab_audio(input_title + '.mid',
                                     soundfont_path=SOUNDFONT_PATH,
                                     sample_rate=16000,
                                     output_for_gradio=True
                                    )
    
    return (16000, midi_audio), midi_plot, input_title + '.mid'

# -----------------------------
# SAVE MIDI FUNCTION
# -----------------------------

def save_midi(tokens, input_title=''):

    time = 0
    dur = 1
    vel = 90
    pitch = 60
    channel = 0
    patch = 0

    patches = [-1] * 16

    channels = [0] * 16
    channels[9] = 1

    song_f = []

    for ss in tokens:

        if 0 <= ss < 256:

            time += ss * 16

        if 256 <= ss < 16768:

            patch = (ss-256) // 128

            if patch < 128:

                if patch not in patches:
                  if 0 in channels:
                      cha = channels.index(0)
                      channels[cha] = 1
                  else:
                      cha = 15

                  patches[cha] = patch
                  channel = patches.index(patch)
                else:
                  channel = patches.index(patch)

            if patch == 128:
                channel = 9

            pitch = (ss-256) % 128


        if 16768 <= ss < 18816:

            dur = ((ss-16768) // 8) * 16
            vel = (((ss-16768) % 8)+1) * 15

            song_f.append(['note', time, dur, channel, pitch, vel, patch])

    patches = [0 if x==-1 else x for x in patches]

    output_score, patches, overflow_patches = TMIDIX.patch_enhanced_score_notes(song_f)
    
    TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(output_score,
                                             output_signature=input_title,
                                             output_file_name=input_title,
                                             track_name='Project Los Angeles',
                                             list_of_MIDI_patches=patches,
                                             verbose=False
                                            )
    
    return output_score

# -----------------------------
# DATASET LOADER FUNCTION
# -----------------------------

def load_dataset(dataset_name):

    print('=' * 70)
    print("Loading requested Orpheus MIDI search dataset...")
    print('=' * 70)

    if dataset_name == 'Multi-Instrumental':
        emb_dataset = hf_hub_download(repo_id=EMB_DATASETS_REPO,
                                      repo_type='dataset',
                                      filename=MI_EMB_DATASET_FILES[0]
                                     )
    
        emb_dataset_corpus = hf_hub_download(repo_id=EMB_DATASETS_REPO,
                                             repo_type='dataset',
                                             filename=MI_EMB_DATASET_FILES[1]
                                            )

    else:
        emb_dataset = hf_hub_download(repo_id=EMB_DATASETS_REPO,
                                      repo_type='dataset',
                                      filename=SP_EMB_DATASET_FILES[0]
                                     )
    
        emb_dataset_corpus = hf_hub_download(repo_id=EMB_DATASETS_REPO,
                                             repo_type='dataset',
                                             filename=SP_EMB_DATASET_FILES[1]
                                            )

    print('=' * 70)
    print("Done!")
    print('=' * 70)
    
    return emb_dataset, emb_dataset_corpus

# -----------------------------
# MAIN MIDI SEARCH FUNCTION
# -----------------------------

def MIDI_Search(title, artist):
    
    """
    Generate tokens using the model, update the composition state, and prepare outputs.
    This function combines seed loading, token generation, and UI output packaging.
    """
    
    print('=' * 70)
    print("Request start time:", datetime.datetime.now(PDT).strftime("%Y-%m-%d %H:%M:%S"))
    start_time = reqtime.time()

    print('=' * 70)
    print('Req title:', title)
    print('Req artist:', artist)
    print('Req dataset:', dataset)
    print('=' * 70)

    if title and artist:
        input_title = title + ' --- ' + artist

    else:
        input_title = ''

        if title:
            input_title = title

        if artist:
            input_title = artist

    print('Searching for best matching title...')
    
    query_embedding = model.encode([input_title])

    similarities = util.cos_sim(query_embedding,
                                embeddings_dataset_corpus
                               )
    
    selected_title_index = np.argmax(similarities).tolist()

    selected_title = song_artist_list[selected_title_index]

    print('Done!')
    print('=' * 70)
    print('Selected title:', selected_title)
    print('Selected title index:', selected_title_index)
    print('=' * 70)
    
    print('Rendering selected title...')
    print('=' * 70)

    final_outputs = []

    song, artist, cap, score, matches = embeddings_dataset[selected_title_index]

    audio, plot, fname = render_midi_output(score, selected_title)

    top_ten_titles_list = [song_artist_list[i] for i, s in matches]

    top_ten_titles = ''

    for i, t in enumerate(top_ten_titles_list):
        top_ten_titles += str(i+1) + ') ' + t + '\n'

    final_outputs.extend([top_ten_titles,
                          selected_title,
                          cap,
                          audio,
                          plot,
                          fname
                         ])

    print('Done!')
    print('=' * 70)
       
    print('Rendering top 10 titles...')
    print('=' * 70)

    for idx, sim in matches:
        
        song, artist, cap, score, matches = embeddings_dataset[idx]
        title = song_artist_list[idx]
    
        audio, plot, fname = render_midi_output(score, title)
    
        final_outputs.extend([title,
                              cap,
                              audio,
                              plot,
                              fname
                             ])
        
    print('Done!')
    print('=' * 70)

    print(top_ten_titles)

    print("Request end time:", datetime.datetime.now(PDT).strftime("%Y-%m-%d %H:%M:%S"))
    print('=' * 70)
    
    end_time = reqtime.time()
    execution_time = end_time - start_time
    
    print(f"Request execution time: {execution_time} seconds")
    print('=' * 70)
        
    return final_outputs

#====================================================================

print('=' * 70)
print('Prepping requested embeddings dataset...')

emb_dat, emb_dat_cor = load_dataset(dataset)

print('=' * 70)
print('Loading requested embeddings dataset...')

embeddings_dataset = TMIDIX.Tegridy_Any_Pickle_File_Reader(emb_dat,
                                                           verbose=False
                                                          )

song_artist_list = [d[0] + ' --- ' + d[1] for d in embeddings_dataset]

embeddings_dataset_corpus = np.load(emb_dat_cor)

print('Done!')
print('=' * 70)

#====================================================================

# -----------------------------
# GRADIO INTERFACE SETUP
# -----------------------------
with gr.Blocks() as demo:

    gr.Markdown("<h1 style='text-align: left; margin-bottom: 1rem'>Orpheus MIDI Search</h1>")
    gr.Markdown("<h1 style='text-align: left; margin-bottom: 1rem'>Search for similar MIDIs with Orpheus embeddings</h1>")

    gr.HTML("""
        Check out <a href="https://huggingface.co/datasets/projectlosangeles/Godzilla-MIDI-Dataset">Godzilla MIDI Dataset</a> on Hugging Face
        <p>
            <a href="https://huggingface.co/spaces/asigalov61/Orpheus-MIDI-Search?duplicate=true">
                <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/duplicate-this-space-md.svg" alt="Duplicate in Hugging Face">
            </a>
        </p>
    """)

    gr.Markdown("# Enter any desired title, artist or both\n\n")
    
    title = gr.Textbox(label="Song Title", value="Family Guy")
    artist = gr.Textbox(label="Song Artist", value="TV Themes")
   
    search_btn = gr.Button(value='Search', variant="primary")
    gr.ClearButton(components=[title, artist])

    gr.Markdown("# Search results")

    gr.Markdown("## Top 10 matches summary")

    top_ten_matches = gr.Textbox(label="Top 10 matches", lines=11, max_lines=11)

    gr.Markdown("## Query MIDI preview")

    query_midi_title = gr.Textbox(label="Query MIDI title", lines=1, max_lines=1)
    query_midi_cap = gr.Textbox(label="Query MIDI caption", lines=7, max_lines=7)
    query_audio = gr.Audio(label="Query MIDI audio", format="wav", elem_id="midi_audio")
    query_plot = gr.Plot(label="Query MIDI score plot")
    query_midi = gr.File(label="Query MIDI file", file_types=[".mid"])

    outputs = [top_ten_matches,
               query_midi_title,
               query_midi_cap,
               query_audio,
               query_plot,
               query_midi
              ]
    
    gr.Markdown("## Top 10 matches previews")
    
    for i in range(10):
        with gr.Tab(f"Match # {i}"):
            title_output = gr.Textbox(label=f"Match # {i} MIDI title", lines=1, max_lines=1)
            cap_output = gr.Textbox(label=f"Match # {i} MIDI caption", lines=7, max_lines=7)
            audio_output = gr.Audio(label=f"Match # {i} MIDI audio", format="mp3")
            plot_output = gr.Plot(label=f"Match # {i} MIDI plot")
            midi_output = gr.File(label=f"Match # {i} MIDI file", file_types=[".mid"])
            
            outputs.extend([title_output,
                            cap_output,
                            audio_output,
                            plot_output,
                            midi_output
                           ])
            
    search_btn.click(
        MIDI_Search,
        [title,
         artist
        ],
        outputs
    )

demo.launch()