Update app.py

app.py

@@ -1,57 +1,42 @@
 import gradio as gr
-import numpy as np
-from librosa import load as librosa_load
-from pydub import AudioSegment
-from pretty_midi import PrettyMIDI
-from transformers import Pop2PianoForConditionalGeneration, Pop2PianoProcessor
-from os import listdir as os_listdir, remove as os_remove
-from io import BytesIO
-from base64 import b64encode
-model = Pop2PianoForConditionalGeneration.from_pretrained("sweetcocoa/pop2piano").to("cpu")
-processor = Pop2PianoProcessor.from_pretrained("sweetcocoa/pop2piano")
-soundfonts = [i.removesuffix(".sf2") for i in os_listdir("soundfonts")]
+from musc.model import PretrainedModel
+from json import load as json_load
+from mido import MidiFile,MidiTrack
+from os import remove as os_remove
+Model = PretrainedModel(json_load(open("violin.json")),"violin_model.pt").to("cpu")
+def merge_violin_tracks(input_midi, output_midi):
+    mid = MidiFile(input_midi)
+    new_mid = MidiFile(ticks_per_beat=mid.ticks_per_beat)
+    new_track = MidiTrack()
+    new_mid.tracks.append(new_track)
+    events = []
+    for track in mid.tracks:
+        current_time = 0
+        for msg in track:
+            current_time += msg.time  
+            events.append((current_time, msg))
+    events.sort(key=lambda x: x[0])
+    last_time = 0
+    for event_time, msg in events:
+        delta_time = event_time - last_time
+        new_track.append(msg.copy(time=delta_time))
+        last_time = event_time  
+    for track in mid.tracks:
+        for msg in track:
+            if msg.type == 'set_tempo':
+                new_track.insert(0, msg)  
+    new_mid.save(output_midi)
 
-def librosa_to_audiosegment(y, sr):
-    epsilon = 1e-8
-    if np.max(np.abs(y)) > 0:
-        y = y / (np.max(np.abs(y)) + epsilon) * 32767
-    return AudioSegment(y.astype(np.int16).tobytes(), frame_rate=sr, sample_width=2, channels=1)
+def transcribe_and_generate_midi(music_file_path, model=Model, batch_size=32):
+    model.transcribe(music_file_path, batch_size=batch_size).write("output.mid")
+    merge_violin_tracks("output.mid","output.mid")
+    os_remove(music_file_path)
+    return "output.mid"
 
-def inference(file_upload, composer, sf2_files):
-    sf2_files = ["soundfonts/" + i + ".sf2" for i in sf2_files]
-    audio_data, audio_sr = librosa_load(file_upload, sr=None)
-    inputs = processor(audio=audio_data, sampling_rate=audio_sr, return_tensors="pt").to("cpu")
-    midi = processor.batch_decode(
-        token_ids=model.generate(input_features=inputs["input_features"], composer="composer" + str(composer)),
-        feature_extractor_output=inputs
-    )["pretty_midi_objects"][0]
-    with open("output.mid", "wb") as midi_file:
-        midi.write(midi_file)
-    midi = PrettyMIDI("output.mid")
-    final_mix = librosa_to_audiosegment(audio_data, audio_sr).apply_gain(-16)
-    for sf2_file in sf2_files:
-        sf_audio_data = midi.fluidsynth(fs=44100, sf2_path=sf2_file)
-        epsilon = 1e-8
-        sf_audio_data = np.int16(sf_audio_data / (np.max(np.abs(sf_audio_data)) + epsilon) * 32767)
-        sf_audio_segment = librosa_to_audiosegment(sf_audio_data, 44100)
-        if len(sf_audio_segment) < len(final_mix):
-            sf_audio_segment = sf_audio_segment.append(AudioSegment.silent(duration=len(final_mix) - len(sf_audio_segment)))
-        elif len(sf_audio_segment) > len(final_mix):
-            sf_audio_segment = sf_audio_segment[:len(final_mix)]
-        final_mix = final_mix.overlay(sf_audio_segment)
-    final_mix.export("output.mp3", format="mp3")
-    os_remove(file_upload)
-    return "output.mid", "output.mp3", f'<div style="display: flex; justify-content: center; align-items: center;"><iframe style="width: 100%; height: 500px; overflow:hidden" srcdoc=\'{open("midi_viz.html").read().replace("{midi_data}", b64encode(open("output.mid","rb").read()).decode("utf-8"))}\'></iframe></div>'
 gr.Interface(
-    inference,
-    [
-        gr.Audio(sources="upload", type="filepath", label="Audio"),
-        gr.Number(1, minimum=1, maximum=21, label="Composer"),
-        gr.Dropdown(soundfonts, multiselect=True, label="Instrument")
-    ],
-    [
-        gr.File(label="MIDI"),
-        gr.Audio(label="Instrument Audio"),
-        gr.HTML()
-    ]
+    fn=transcribe_and_generate_midi, 
+    inputs=gr.Audio(label="Upload your Audio file",type="filepath"),
+    outputs=gr.File(label="Download MIDI file"),
+    title="Audio2Violin",
+    description="Upload a Audio file, and it will be transcribed into Violin MIDI format."
 ).launch()