audio2midi 0.5.0__py2.py3-none-any.whl → 0.7.0__py2.py3-none-any.whl

audio2midi/librosa_pitch_detector.py

@@ -147,6 +147,6 @@ class Guitar_Pitch_Det:
                 # If it's a rest, just advance the time
                 duration_sec = duration * (60 / tempo_bpm)
                 note_time += duration_sec
-        pm.instruments.append(piano)
+        pm.instruments.append(instrument)
         pm.write(output_file)
         return output_file

audio2midi/magenta_music_transcription.py

@@ -0,0 +1,76 @@
+from typing import Callable
+import numpy as np
+import pretty_midi_fix
+import tensorflow
+import librosa.feature.rhythm
+from huggingface_hub import hf_hub_download
+import librosa
+
+
+def endpitch(pitch, endFrame,volProb,intervals,MIN_NOTE_FRAMES,frameLenSecs,PITCH_BEND_ENABLED,pitch_bend_steps,pitch_bend_depth,track):
+    startFrame = intervals[pitch]
+    if endFrame - startFrame < MIN_NOTE_FRAMES:
+        return
+    vol = volProb[startFrame, pitch]
+    if vol < 0 or vol > 1:
+        return
+    start_time = startFrame * frameLenSecs
+    track.notes.append(pretty_midi_fix.Note(velocity=int(max(0, min(1, vol)) * 80 + 10), pitch=pitch + 21, start=start_time, end=endFrame * frameLenSecs))
+    if PITCH_BEND_ENABLED:
+        for step in range(pitch_bend_steps):
+            track.pitch_bends.append(pretty_midi_fix.PitchBend(pitch=int(np.sin(np.pi * step / (pitch_bend_steps - 1)) * pitch_bend_depth), time=start_time + step * 0.01))
+        track.pitch_bends.append(pretty_midi_fix.PitchBend(pitch=0, time=start_time + 0.05))  # Reset
+    del intervals[pitch]
+
+def model_output_to_notes(model_output,onset_thresh,include_pitch_bends,min_note_len,gap_tolerance_frames,pitch_bend_depth,pitch_bend_steps):
+    actProb , onProb , offProb , volProb , tempo = model_output
+    midi = pretty_midi_fix.PrettyMIDI(initial_tempo=tempo)
+    track = pretty_midi_fix.Instrument(program=40)
+    frameLenSecs = librosa.frames_to_time(1, sr=16000)
+    intervals = {}
+    onsets = (onProb > onset_thresh).astype(np.int8)
+    frames = onsets | (actProb > onset_thresh).astype(np.int8)
+    for i, frame in enumerate(np.vstack([frames, np.zeros(frames.shape[1])])):
+        for pitch, active in enumerate(frame):
+            if active:
+                if pitch not in intervals:
+                    if onsets is None or onsets[i, pitch]:
+                        intervals[pitch] = i
+                elif onsets is not None and onsets[i, pitch] and (i - intervals[pitch] > 2):
+                    endpitch(pitch, i,volProb,intervals,min_note_len,frameLenSecs,include_pitch_bends,pitch_bend_steps,pitch_bend_depth,track)
+                    intervals[pitch] = i
+            elif pitch in intervals:
+                if i + gap_tolerance_frames < frames.shape[0] and np.any(frames[i:i + gap_tolerance_frames, pitch]):
+                    continue  # Don't end the note yet
+                endpitch(pitch, i,volProb,intervals,min_note_len,frameLenSecs,include_pitch_bends,pitch_bend_steps,pitch_bend_depth,track)
+    midi.instruments.append(track)
+    return midi
+
+class Magenta:
+    def __init__(self,model_path=hf_hub_download("shethjenil/Audio2Midi_Models","magenta.tflite")):
+        self.interp = tensorflow.lite.Interpreter(model_path=model_path)
+        self.interp.allocate_tensors()
+        self.inputLen = self.interp.get_input_details()[0]['shape'][0]
+        self.outputStep = self.interp.get_output_details()[0]['shape'][1] * 512
+
+    def run_inference(self,audio_path,progress_callback):
+        song = librosa.load(audio_path,sr=16000)[0]
+        actProb, onProb, offProb, volProb = np.empty((1, 88)), np.empty((1, 88)), np.empty((1, 88)), np.empty((1, 88))
+        paddedSong = np.append(song, np.zeros(-(song.size - self.inputLen) % self.outputStep, dtype=np.float32))
+        total_size = (paddedSong.size - self.inputLen) // self.outputStep + 1
+        tempo = librosa.feature.rhythm.tempo(y=song, sr=16000).mean()
+        for i in range(total_size):
+            self.interp.set_tensor(self.interp.get_input_details()[0]['index'], paddedSong[i * self.outputStep : i * self.outputStep + self.inputLen])
+            self.interp.invoke()
+            actProb = np.vstack((actProb, self.interp.get_tensor(self.interp.get_output_details()[0]['index'])[0]))
+            onProb  = np.vstack(( onProb, self.interp.get_tensor(self.interp.get_output_details()[1]['index'])[0]))
+            offProb = np.vstack((offProb, self.interp.get_tensor(self.interp.get_output_details()[2]['index'])[0]))
+            volProb = np.vstack((volProb, self.interp.get_tensor(self.interp.get_output_details()[3]['index'])[0]))
+            if progress_callback:
+                progress_callback(i,total_size)
+        return actProb , onProb , offProb , volProb , tempo
+
+
+    def predict(self,audio,onset_thresh=0,min_note_len=3,include_pitch_bends=True,gap_tolerance_frames = 3,pitch_bend_depth = 1500,pitch_bend_steps = 4,progress_callback: Callable[[int, int], None] = None,output_file="output.mid"):
+        model_output_to_notes(self.run_inference(audio,progress_callback),onset_thresh  = onset_thresh,min_note_len  = min_note_len,include_pitch_bends  = include_pitch_bends,pitch_bend_depth=pitch_bend_depth,pitch_bend_steps=pitch_bend_steps,gap_tolerance_frames=gap_tolerance_frames).write(output_file)
+        return output_file

audio2midi/melodia_pitch_detector.py

@@ -49,10 +49,11 @@ class Melodia():
         data, sr = librosa_load(audio, sr=44100, mono=True)
         pm = PrettyMIDI(initial_tempo=tempo)
         instrument = Instrument(program=40)
-        seconds_per_beat = 60.0 / tempo
-        for onset_beats, duration_beats, pitch in self.midi_to_notes(self.hz2midi(np.insert(vamp_collect(data, sr, "mtg-melodia:melodia",parameters={"voicing": 0.2})['vector'][1], 0, [0]*8)), 44100, smooth, minduration,hop):
-            start = onset_beats * seconds_per_beat
-            instrument.notes.append(Note(100,int(pitch),start,start + (duration_beats * seconds_per_beat)))
+        for onset_sec, duration_sec, pitch in self.midi_to_notes(
+                self.hz2midi(np.insert(vamp_collect(data, sr, "mtg-melodia:melodia", parameters={"voicing": 0.2})['vector'][1],0, [0]*8)), 44100, smooth, minduration, hop):
+            start = onset_sec
+            end = start + duration_sec
+            instrument.notes.append(Note(100, int(pitch), start, end))
         pm.instruments.append(instrument)
         pm.write(output_file)
         return output_file

{audio2midi-0.5.0.dist-info → audio2midi-0.7.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: audio2midi
-Version: 0.5.0
+Version: 0.7.0
 Summary: Audio To Midi
 Author-email: dummyjenil <dummyjenil@gmail.com>
 Provides-Extra: all
@@ -52,6 +52,12 @@ Provides-Extra: librosa-pitch-detector
 Requires-Dist: librosa; extra == 'librosa-pitch-detector'
 Requires-Dist: numpy; extra == 'librosa-pitch-detector'
 Requires-Dist: pretty-midi-fix; extra == 'librosa-pitch-detector'
+Provides-Extra: magenta-music-transcription
+Requires-Dist: huggingface-hub; extra == 'magenta-music-transcription'
+Requires-Dist: librosa; extra == 'magenta-music-transcription'
+Requires-Dist: numpy; extra == 'magenta-music-transcription'
+Requires-Dist: pretty-midi-fix; extra == 'magenta-music-transcription'
+Requires-Dist: tensorflow; extra == 'magenta-music-transcription'
 Provides-Extra: melodia-pitch-detector
 Requires-Dist: huggingface-hub; extra == 'melodia-pitch-detector'
 Requires-Dist: librosa; extra == 'melodia-pitch-detector'

{audio2midi-0.5.0.dist-info → audio2midi-0.7.0.dist-info}/RECORD

@@ -2,12 +2,13 @@ audio2midi/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 audio2midi/basic_pitch_pitch_detector.py,sha256=6ZH4SKL9qAj6ODSr84qr_QmNvDYOuZRUrbVCOB9430E,30663
 audio2midi/crepe_pitch_detector.py,sha256=63PLW5pQ_WSlIr_bOSAv7wjIujA06Iy-pTQab3InPR4,34675
 audio2midi/crepe_pitch_detector_tf.py,sha256=obOK9zWeAFWKhVuZAXG-SM2Cn-vVBjQMU08hwwjDvKA,7607
-audio2midi/librosa_pitch_detector.py,sha256=btwYkNqVgkf9rdkBtH2Q5DqdNVKtO9iZ88SvcbQK1jk,7777
-audio2midi/melodia_pitch_detector.py,sha256=YFt9NKyZ_Dyt_3ltInUz6QZgovmmPK6gR9EPlL_aV5Y,2402
+audio2midi/librosa_pitch_detector.py,sha256=U8r98mnkkRxTlG2M0fX5Fif3qwihgfNry1pa-4lEd-E,7782
+audio2midi/magenta_music_transcription.py,sha256=7a8OAqO3oGbQnXGC8DZ8qUpCOO7Ho4v-ATxIW7kF3XE,4878
+audio2midi/melodia_pitch_detector.py,sha256=fYueHUjDeXXY4Aea5aOcJRT6E66NZkHYfG3D5fofYnw,2358
 audio2midi/mt3_music_transcription.py,sha256=_8MgMoXd-LNzI4GeXwIJs6pq7B7lrEkzOQwQSPTo1wo,32
 audio2midi/pop2piano.py,sha256=kBAF1kY-5Ctu92etNLo4Clr1hkW1B5OCvd-XT7SrI8g,125685
 audio2midi/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 audio2midi/violin_pitch_detector.py,sha256=C0--R0NaYcIyAFELGgebbs3yjAnQeVdtpk4iZ2NbUIA,61650
-audio2midi-0.5.0.dist-info/METADATA,sha256=V9B8lUncLjTxdWvM25lY2RCTTmqTwagsEOrjGP2o7z8,12857
-audio2midi-0.5.0.dist-info/WHEEL,sha256=tkmg4JIqwd9H8mL30xA7crRmoStyCtGp0VWshokd1Jc,105
-audio2midi-0.5.0.dist-info/RECORD,,
+audio2midi-0.7.0.dist-info/METADATA,sha256=k7Uk8cWS26SouSyi6o3x9o5Sr6DJn85AbL-86rUJhRY,13233
+audio2midi-0.7.0.dist-info/WHEEL,sha256=tkmg4JIqwd9H8mL30xA7crRmoStyCtGp0VWshokd1Jc,105
+audio2midi-0.7.0.dist-info/RECORD,,