add asigalov61/Music-Llama tv2o-large model

app.py

@@ -1,14 +1,13 @@
 import argparse
 import glob
+import json
 import os.path
 import time
-import uuid
 
 import gradio as gr
 import numpy as np
 import onnxruntime as rt
 import tqdm
-import json
 from huggingface_hub import hf_hub_download
 
 import MIDI
@@ -47,6 +46,7 @@ def sample_top_p_k(probs, p, k, generator=None):
 
 def generate(model, prompt=None, max_len=512, temp=1.0, top_p=0.98, top_k=20,
  disable_patch_change=False, disable_control_change=False, disable_channels=None, generator=None):
+ tokenizer = model[2]
  if disable_channels is not None:
  disable_channels = [tokenizer.parameter_ids["channel"][c] for c in disable_channels]
  else:
@@ -121,10 +121,11 @@ def send_msgs(msgs):
  return json.dumps(msgs)
 
 
-def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events,
+def run(model_name, tab, mid_seq, instruments, drum_kit, bpm, mid, midi_events,
  reduce_cc_st, remap_track_channel, add_default_instr, remove_empty_channels, seed, seed_rand,
  gen_events, temp, top_p, top_k, allow_cc):
- mid_seq = []
+ model = models[model_name]
+ tokenizer = model[2]
  bpm = int(bpm)
  gen_events = int(gen_events)
  max_len = gen_events
@@ -137,7 +138,7 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events,
  i = 0
  mid = [[tokenizer.bos_id] + [tokenizer.pad_id] * (tokenizer.max_token_seq - 1)]
  if bpm != 0:
- mid.append(tokenizer.event2tokens(["set_tempo",0,0,0, bpm]))
+ mid.append(tokenizer.event2tokens(["set_tempo", 0, 0, 0, bpm]))
  patches = {}
  if instruments is None:
  instruments = []
@@ -153,7 +154,7 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events,
  if len(instruments) > 0:
  disable_patch_change = True
  disable_channels = [i for i in range(16) if i not in patches]
- elif mid is not None:
+ elif tab == 1 and mid is not None:
  eps = 4 if reduce_cc_st else 0
  mid = tokenizer.tokenize(MIDI.midi2score(mid), cc_eps=eps, tempo_eps=eps,
  remap_track_channel=remap_track_channel,
@@ -161,14 +162,26 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events,
  remove_empty_channels=remove_empty_channels)
  mid = np.asarray(mid, dtype=np.int64)
  mid = mid[:int(midi_events)]
+ mid_seq = []
  for token_seq in mid:
  mid_seq.append(token_seq.tolist())
- max_len += len(mid)
+ elif tab == 2 and mid_seq is not None:
+ mid = np.asarray(mid_seq, dtype=np.int64)
+ else:
+ mid_seq = []
+ mid = None
+
+ if mid is not None:
+ max_len += len(mid)
+
  events = [tokenizer.tokens2event(tokens) for tokens in mid_seq]
- init_msgs = [create_msg("visualizer_clear", None), create_msg("visualizer_append", events)]
- t = time.time() + 1
+ if tab == 2:
+ init_msgs = [create_msg("visualizer_continue", tokenizer.version)]
+ else:
+ init_msgs = [create_msg("visualizer_clear", tokenizer.version),
+ create_msg("visualizer_append", events)]
  yield mid_seq, None, None, seed, send_msgs(init_msgs)
- model = models[model_name]
+ t = time.time() + 1
  midi_generator = generate(model, mid, max_len=max_len, temp=temp, top_p=top_p, top_k=top_k,
  disable_patch_change=disable_patch_change, disable_control_change=not allow_cc,
  disable_channels=disable_channels, generator=generator)
@@ -191,9 +204,10 @@ def run(model_name, tab, instruments, drum_kit, bpm, mid, midi_events,
  yield mid_seq, "output.mid", (44100, audio), seed, send_msgs([create_msg("visualizer_end", events)])
 
 
-def cancel_run(mid_seq):
+def cancel_run(model_name, mid_seq):
  if mid_seq is None:
  return None, None, []
+ tokenizer = models[model_name][2]
  mid = tokenizer.detokenize(mid_seq)
  with open(f"output.mid", 'wb') as f:
  f.write(MIDI.score2midi(mid))
@@ -233,6 +247,20 @@ def hf_hub_download_retry(repo_id, filename):
  if err:
  raise err
 
+def get_tokenizer(config_name):
+ tv, size = config_name.split("-")
+ tv = tv[1:]
+ if tv[-1] == "o":
+ o = True
+ tv = tv[:-1]
+ else:
+ o = False
+ if tv not in ["v1", "v2"]:
+ raise ValueError(f"Unknown tokenizer version {tv}")
+ tokenizer = MIDITokenizer(tv)
+ tokenizer.set_optimise_midi(o)
+ return tokenizer
+
 number2drum_kits = {-1: "None", 0: "Standard", 8: "Room", 16: "Power", 24: "Electric", 25: "TR-808", 32: "Jazz",
  40: "Blush", 48: "Orchestra"}
 patch2number = {v: k for k, v in MIDI.Number2patch.items()}
@@ -245,19 +273,20 @@ if __name__ == "__main__":
  parser.add_argument("--max-gen", type=int, default=1024, help="max")
  opt = parser.parse_args()
  soundfont_path = hf_hub_download_retry(repo_id="skytnt/midi-model", filename="soundfont.sf2")
- models_info = {"generic pretrain model": ["skytnt/midi-model", ""],
- "j-pop finetune model": ["skytnt/midi-model-ft", "jpop/"],
- "touhou finetune model": ["skytnt/midi-model-ft", "touhou/"],
+ models_info = {"generic pretrain model (tv2o-large)": ["asigalov61/Music-Llama", "", "tv2o-large"],
+ "generic pretrain model (tv1-medium)": ["skytnt/midi-model", "", "tv1-medium"],
+ "j-pop finetune model (tv1-medium)": ["skytnt/midi-model-ft", "jpop/", "tv1-medium"],
+ "touhou finetune model (tv1-medium)": ["skytnt/midi-model-ft", "touhou/", "tv1-medium"],
  }
  models = {}
- tokenizer = MIDITokenizer()
  providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
- for name, (repo_id, path) in models_info.items():
+ for name, (repo_id, path, config) in models_info.items():
  model_base_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_base.onnx")
  model_token_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_token.onnx")
  model_base = rt.InferenceSession(model_base_path, providers=providers)
  model_token = rt.InferenceSession(model_token_path, providers=providers)
- models[name] = [model_base, model_token]
+ tokenizer = get_tokenizer(config)
+ models[name] = [model_base, model_token, tokenizer]
 
  load_javascript()
  app = gr.Blocks()
@@ -316,9 +345,12 @@ if __name__ == "__main__":
  input_remove_empty_channels = gr.Checkbox(label="remove channels without notes", value=False)
  example2 = gr.Examples([[file, 128] for file in glob.glob("example/*.mid")],
  [input_midi, input_midi_events])
+ with gr.TabItem("last output prompt") as tab3:
+ gr.Markdown("Continue generating on the last output. Just click the generate button")
 
  tab1.select(lambda: 0, None, tab_select, queue=False)
  tab2.select(lambda: 1, None, tab_select, queue=False)
+ tab3.select(lambda: 2, None, tab_select, queue=False)
  input_seed = gr.Slider(label="seed", minimum=0, maximum=2 ** 31 - 1,
  step=1, value=0)
  input_seed_rand = gr.Checkbox(label="random seed", value=True)
@@ -336,12 +368,14 @@ if __name__ == "__main__":
  output_midi_visualizer = gr.HTML(elem_id="midi_visualizer_container")
  output_audio = gr.Audio(label="output audio", format="mp3", elem_id="midi_audio")
  output_midi = gr.File(label="output midi", file_types=[".mid"])
- run_event = run_btn.click(run, [input_model, tab_select, input_instruments, input_drum_kit, input_bpm,
- input_midi, input_midi_events, input_reduce_cc_st, input_remap_track_channel,
- input_add_default_instr, input_remove_empty_channels, input_seed,
- input_seed_rand, input_gen_events, input_temp, input_top_p, input_top_k,
- input_allow_cc],
+ run_event = run_btn.click(run, [input_model, tab_select, output_midi_seq, input_instruments,
+ input_drum_kit, input_bpm, input_midi, input_midi_events, input_reduce_cc_st,
+ input_remap_track_channel, input_add_default_instr, input_remove_empty_channels,
+ input_seed, input_seed_rand, input_gen_events, input_temp, input_top_p,
+ input_top_k, input_allow_cc],
  [output_midi_seq, output_midi, output_audio, input_seed, js_msg],
  concurrency_limit=3)
- stop_btn.click(cancel_run, [output_midi_seq], [output_midi, output_audio, js_msg], cancels=run_event, queue=False)
+ stop_btn.click(cancel_run, [input_model,output_midi_seq],
+ [output_midi, output_audio, js_msg],
+ cancels=run_event, queue=False)
  app.launch(server_port=opt.port, share=opt.share, inbrowser=True)

javascript/app.js

@@ -117,9 +117,11 @@ class MidiVisualizer extends HTMLElement{
  this.totalTimeMs = 0
  this.playTime = 0
  this.playTimeMs = 0
+ this.lastUpdateTime = 0
  this.colorMap = new Map();
  this.playing = false;
  this.timer = null;
+ this.version = "v2"
  this.init();
  }
 
@@ -157,6 +159,7 @@ class MidiVisualizer extends HTMLElement{
  this.setPlayTime(0);
  this.totalTimeMs = 0;
  this.playTimeMs = 0
+ this.lastUpdateTime = 0
  this.svgWidth = 0
  this.svg.innerHTML = ''
  this.svg.style.width = `${this.svgWidth}px`;
@@ -171,10 +174,22 @@ class MidiVisualizer extends HTMLElement{
  midiEvent = [midiEvent[0], t].concat(midiEvent.slice(3))
  if(midiEvent[0] === "note"){
  let track = midiEvent[2]
- let duration = midiEvent[3]
- let channel = midiEvent[4]
- let pitch = midiEvent[5]
- let velocity = midiEvent[6]
+ let duration = 0
+ let channel = 0
+ let pitch = 0
+ let velocity = 0
+ if(this.version === "v1"){
+ duration = midiEvent[3]
+ channel = midiEvent[4]
+ pitch = midiEvent[5]
+ velocity = midiEvent[6]
+ }else if (this.version === "v2"){
+ channel = midiEvent[3]
+ pitch = midiEvent[4]
+ velocity = midiEvent[5]
+ duration = midiEvent[6]
+ }
+
  let x = (t/this.timePreBeat)*this.config.beatWidth
  let y = (127 - pitch)*this.config.noteHeight
  let w = (duration/this.timePreBeat)*this.config.beatWidth
@@ -252,14 +267,14 @@ class MidiVisualizer extends HTMLElement{
  this.timeLine.setAttribute('y2', `${this.config.noteHeight*128}`);
 
  this.wrapper.scrollTo(Math.max(0, x - this.wrapper.offsetWidth/2), 0)
-
- if(this.playing){
+ let dt = Date.now() - this.lastUpdateTime; // limit the update rate of ActiveNotes
+ if(this.playing && dt > 50){
  let activeNotes = []
  this.removeActiveNotes(this.activeNotes)
  this.midiEvents.forEach((midiEvent)=>{
  if(midiEvent[0] === "note"){
  let time = midiEvent[1]
- let duration = midiEvent[3]
+ let duration = this.version==="v1"? midiEvent[3]:midiEvent[6]
  let note = midiEvent[midiEvent.length - 1]
  if(time <=this.playTime && time+duration>= this.playTime){
  activeNotes.push(note)
@@ -267,7 +282,9 @@ class MidiVisualizer extends HTMLElement{
  }
  })
  this.addActiveNotes(activeNotes)
+ this.lastUpdateTime = Date.now();
  }
+
  }
 
  setPlayTimeMs(ms){
@@ -424,6 +441,11 @@ customElements.define('midi-visualizer', MidiVisualizer);
  switch (msg.name) {
  case "visualizer_clear":
  midi_visualizer.clearMidiEvents(false);
+ midi_visualizer.version = msg.data
+ createProgressBar(midi_visualizer_container_inited)
+ break;
+ case "visualizer_continue":
+ midi_visualizer.version = msg.data
  createProgressBar(midi_visualizer_container_inited)
  break;
  case "visualizer_append":

midi_tokenizer.py

@@ -1,11 +1,13 @@
 import random
 
-import PIL
+import PIL.Image
 import numpy as np
 
 
-class MIDITokenizer:
+class MIDITokenizerV1:
  def __init__(self):
+ self.version = "v1"
+ self.optimise_midi = False
  self.vocab_size = 0
 
  def allocate_ids(size):
@@ -31,26 +33,38 @@ class MIDITokenizer:
  self.parameter_ids = {p: allocate_ids(s) for p, s in self.event_parameters.items()}
  self.max_token_seq = max([len(ps) for ps in self.events.values()]) + 1
 
- def tempo2bpm(self, tempo):
+ def set_optimise_midi(self, optimise_midi=True):
+ self.optimise_midi = optimise_midi
+
+ @staticmethod
+ def tempo2bpm(tempo):
  tempo = tempo / 10 ** 6 # us to s
  bpm = 60 / tempo
  return bpm
 
- def bpm2tempo(self, bpm):
+ @staticmethod
+ def bpm2tempo(bpm):
  if bpm == 0:
  bpm = 1
  tempo = int((60 / bpm) * 10 ** 6)
  return tempo
 
  def tokenize(self, midi_score, add_bos_eos=True, cc_eps=4, tempo_eps=4,
- remap_track_channel=False, add_default_instr=False, remove_empty_channels=False):
+ remap_track_channel=None, add_default_instr=None, remove_empty_channels=None):
+ if remap_track_channel is None: # set default value
+ remap_track_channel = self.optimise_midi
+ if add_default_instr is None:
+ add_default_instr = self.optimise_midi
+ if remove_empty_channels is None:
+ remove_empty_channels = self.optimise_midi
+
  ticks_per_beat = midi_score[0]
  event_list = {}
  track_idx_map = {i: dict() for i in range(16)}
  track_idx_dict = {}
  channels = []
  patch_channels = []
- empty_channels = [True]*16
+ empty_channels = [True] * 16
  channel_note_tracks = {i: list() for i in range(16)}
  for track_idx, track in enumerate(midi_score[1:129]):
  last_notes = {}
@@ -74,7 +88,7 @@ class MIDITokenizer:
  note_tracks.append(track_idx)
  new_event[4] = max(1, round(16 * new_event[4] / ticks_per_beat))
  elif event[0] == "set_tempo":
- if new_event[4] == 0: # invalid tempo
+ if new_event[4] == 0: # invalid tempo
  continue
  bpm = int(self.tempo2bpm(new_event[4]))
  new_event[4] = min(bpm, 255)
@@ -143,8 +157,8 @@ class MIDITokenizer:
  channels = list(channels_map.values())
 
  track_count = 0
- track_idx_map_order = [k for k,v in sorted(list(channels_map.items()), key=lambda x: x[1])]
- for c in track_idx_map_order: # tracks not to remove
+ track_idx_map_order = [k for k, v in sorted(list(channels_map.items()), key=lambda x: x[1])]
+ for c in track_idx_map_order: # tracks not to remove
  if remove_empty_channels and c in empty_channels:
  continue
  tr_map = track_idx_map[c]
@@ -154,7 +168,7 @@ class MIDITokenizer:
  continue
  track_count += 1
  tr_map[track_idx] = track_count
- for c in track_idx_map_order: # tracks to remove
+ for c in track_idx_map_order: # tracks to remove
  if not (remove_empty_channels and c in empty_channels):
  continue
  tr_map = track_idx_map[c]
@@ -166,7 +180,7 @@ class MIDITokenizer:
  tr_map[track_idx] = track_count
 
  empty_channels = [channels_map[c] for c in empty_channels]
-
+ track_idx_dict = {}
  for event in event_list:
  name = event[0]
  track_idx = event[3]
@@ -174,7 +188,8 @@ class MIDITokenizer:
  c = event[5]
  event[5] = channels_map[c]
  event[3] = track_idx_map[c][track_idx]
- track_idx_dict[event[5]] = event[3]
+ track_idx_dict.setdefault(event[5], event[3])
+ # setdefault, so the track_idx is first of the channel
  elif name == "set_tempo":
  event[3] = 0
  elif name == "control_change" or name == "patch_change":
@@ -192,10 +207,10 @@ class MIDITokenizer:
 
  if add_default_instr:
  for c in channels:
- if c not in patch_channels:
- event_list.append(["patch_change", 0,0, track_idx_dict[c], c, 0])
+ if c not in patch_channels and c in track_idx_dict:
+ event_list.append(["patch_change", 0, 0, track_idx_dict[c], c, 0])
 
- events_name_order = {"set_tempo":0, "patch_change":1, "control_change":2, "note":3}
+ events_name_order = {"set_tempo": 0, "patch_change": 1, "control_change": 2, "note": 3}
  events_order = lambda e: e[1:4] + [events_name_order[e[0]]]
  event_list = sorted(event_list, key=events_order)
 
@@ -214,7 +229,7 @@ class MIDITokenizer:
  if notes_in_setup and i > 0:
  pre_event = event_list[i - 1]
  has_pre = event[1] + event[2] == pre_event[1] + pre_event[2]
- if (event[0] == "note" and not has_next) or (notes_in_setup and not has_pre) :
+ if (event[0] == "note" and not has_next) or (notes_in_setup and not has_pre):
  event_list = sorted(setup_events.values(), key=events_order) + event_list[i:]
  break
  else:
@@ -253,17 +268,17 @@ class MIDITokenizer:
  return tokens
 
  def tokens2event(self, tokens):
- if tokens[0] in self.id_events:
- name = self.id_events[tokens[0]]
- if len(tokens) <= len(self.events[name]):
- return []
- params = tokens[1:]
- params = [params[i] - self.parameter_ids[p][0] for i, p in enumerate(self.events[name])]
- if not all([0 <= params[i] < self.event_parameters[p] for i, p in enumerate(self.events[name])]):
- return []
- event = [name] + params
- return event
- return []
+ if tokens[0] not in self.id_events:
+ return []
+ name = self.id_events[tokens[0]]
+ if len(tokens) <= len(self.events[name]):
+ return []
+ params = tokens[1:]
+ params = [params[i] - self.parameter_ids[p][0] for i, p in enumerate(self.events[name])]
+ if not all([0 <= params[i] < self.event_parameters[p] for i, p in enumerate(self.events[name])]):
+ return []
+ event = [name] + params
+ return event
 
  def detokenize(self, midi_seq):
  ticks_per_beat = 480
@@ -386,7 +401,9 @@ class MIDITokenizer:
  midi_seq_new.append(tokens_new)
  return midi_seq_new
 
- def check_quality(self, midi_seq, alignment_min=0.4, tonality_min=0.8, piano_max=0.7, notes_bandwidth_min=3, notes_density_max=30, notes_density_min=2.5, total_notes_max=10000, total_notes_min=500, note_window_size=16):
+ def check_quality(self, midi_seq, alignment_min=0.3, tonality_min=0.8, piano_max=0.7, notes_bandwidth_min=3,
+ notes_density_max=50, notes_density_min=2.5, total_notes_max=20000, total_notes_min=256,
+ note_window_size=16):
  total_notes = 0
  channels = []
  time_hist = [0] * 16
@@ -450,13 +467,648 @@ class MIDITokenizer:
  tonality_list.append(sum(key_hist[:7]) / len(notes))
  notes_density_list.append(len(notes) / note_window_size)
  tonality_list = sorted(tonality_list)
- tonality = sum(tonality_list)/len(tonality_list)
- notes_bandwidth = sum(notes_bandwidth_list)/len(notes_bandwidth_list) if notes_bandwidth_list else 0
+ tonality = sum(tonality_list) / len(tonality_list)
+ notes_bandwidth = sum(notes_bandwidth_list) / len(notes_bandwidth_list) if notes_bandwidth_list else 0
+ notes_density = max(notes_density_list) if notes_density_list else 0
+ piano_ratio = len(piano_channels) / len(channels)
+ if len(channels) <= 3: # ignore piano threshold if it is a piano solo midi
+ piano_max = 1
+ if alignment < alignment_min: # check weather the notes align to the bars (because some midi files are recorded)
+ reasons.append("alignment")
+ if tonality < tonality_min: # check whether the music is tonal
+ reasons.append("tonality")
+ if notes_bandwidth < notes_bandwidth_min: # check whether music is melodic line only
+ reasons.append("bandwidth")
+ if not notes_density_min < notes_density < notes_density_max:
+ reasons.append("density")
+ if piano_ratio > piano_max: # check whether most instruments is piano (because some midi files don't have instruments assigned correctly)
+ reasons.append("piano")
+ return not reasons, reasons
+
+
+class MIDITokenizerV2:
+ def __init__(self):
+ self.version = "v2"
+ self.optimise_midi = False
+ self.vocab_size = 0
+
+ def allocate_ids(size):
+ ids = [self.vocab_size + i for i in range(size)]
+ self.vocab_size += size
+ return ids
+
+ self.pad_id = allocate_ids(1)[0]
+ self.bos_id = allocate_ids(1)[0]
+ self.eos_id = allocate_ids(1)[0]
+ self.events = {
+ "note": ["time1", "time2", "track", "channel", "pitch", "velocity", "duration"],
+ "patch_change": ["time1", "time2", "track", "channel", "patch"],
+ "control_change": ["time1", "time2", "track", "channel", "controller", "value"],
+ "set_tempo": ["time1", "time2", "track", "bpm"],
+ "time_signature": ["time1", "time2", "track", "nn", "dd"],
+ "key_signature": ["time1", "time2", "track", "sf", "mi"],
+ }
+ self.event_parameters = {
+ "time1": 128, "time2": 16, "duration": 2048, "track": 128, "channel": 16, "pitch": 128, "velocity": 128,
+ "patch": 128, "controller": 128, "value": 128, "bpm": 384, "nn": 16, "dd": 4, "sf": 15, "mi": 2
+ }
+ self.event_ids = {e: allocate_ids(1)[0] for e in self.events.keys()}
+ self.id_events = {i: e for e, i in self.event_ids.items()}
+ self.parameter_ids = {p: allocate_ids(s) for p, s in self.event_parameters.items()}
+ self.max_token_seq = max([len(ps) for ps in self.events.values()]) + 1
+
+ def set_optimise_midi(self, optimise_midi=True):
+ self.optimise_midi = optimise_midi
+
+ @staticmethod
+ def tempo2bpm(tempo):
+ tempo = tempo / 10 ** 6 # us to s
+ bpm = 60 / tempo
+ return bpm
+
+ @staticmethod
+ def bpm2tempo(bpm):
+ if bpm == 0:
+ bpm = 1
+ tempo = int((60 / bpm) * 10 ** 6)
+ return tempo
+
+ @staticmethod
+ def sf2key(sf):
+ # sf in key_signature to key.
+ # key represents the sequence from C note to B note (12 in total)
+ return (sf * 7) % 12
+
+ @staticmethod
+ def key2sf(k, mi):
+ # key to sf
+ sf = (k * 7) % 12
+ if sf > 6 or (mi == 1 and sf >= 5):
+ sf -= 12
+ return sf
+
+ @staticmethod
+ def detect_key_signature(key_hist, threshold=0.7):
+ if len(key_hist) != 12:
+ return None
+ p = sum(sorted(key_hist, reverse=True)[:7]) / sum(key_hist)
+ if p < threshold:
+ return None
+ keys = [x[1] for x in sorted(zip(key_hist, range(len(key_hist))), reverse=True, key=lambda x: x[0])[:7]]
+ keys = sorted(keys)
+ semitones = []
+ for i in range(len(keys)):
+ dis = keys[i] - keys[i - 1]
+ if dis == 1 or dis == -11:
+ semitones.append(keys[i])
+ if len(semitones) != 2:
+ return None
+ semitones_dis = semitones[1] - semitones[0]
+ if semitones_dis == 5:
+ root_key = semitones[0]
+ elif semitones_dis == 7:
+ root_key = semitones[1]
+ else:
+ return None
+ return root_key
+
+ def tokenize(self, midi_score, add_bos_eos=True, cc_eps=4, tempo_eps=4,
+ remap_track_channel=None, add_default_instr=None, remove_empty_channels=None):
+ if remap_track_channel is None: # set default value
+ remap_track_channel = self.optimise_midi
+ if add_default_instr is None:
+ add_default_instr = self.optimise_midi
+ if remove_empty_channels is None:
+ remove_empty_channels = self.optimise_midi
+
+ ticks_per_beat = midi_score[0]
+ event_list = {}
+ track_idx_map = {i: dict() for i in range(16)}
+ track_idx_dict = {}
+ channels = []
+ patch_channels = []
+ empty_channels = [True] * 16
+ channel_note_tracks = {i: list() for i in range(16)}
+ note_key_hist = [0]*12
+ key_sig_num = 0
+ track_to_channels = {}
+ for track_idx, track in enumerate(midi_score[1:129]):
+ last_notes = {}
+ patch_dict = {}
+ control_dict = {}
+ last_bpm = 0
+ track_channels = []
+ track_to_channels.setdefault(track_idx, track_channels)
+ for event in track:
+ if event[0] not in self.events:
+ continue
+ name = event[0]
+ c = -1
+ t = round(16 * event[1] / ticks_per_beat) # quantization
+ new_event = [name, t // 16, t % 16, track_idx]
+ if name == "note":
+ d, c, p, v = event[2:]
+ if not (0 <= c <= 15):
+ continue
+ d = max(1, round(16 * d / ticks_per_beat))
+ new_event += [c, p, v, d]
+ empty_channels[c] = False
+ track_idx_dict.setdefault(c, track_idx)
+ note_tracks = channel_note_tracks[c]
+ if track_idx not in note_tracks:
+ note_tracks.append(track_idx)
+ if c != 9:
+ note_key_hist[p%12] += 1
+ if c not in track_channels:
+ track_channels.append(c)
+ elif name == "patch_change":
+ c, p = event[2:]
+ if not (0 <= c <= 15):
+ continue
+ new_event += [c, p]
+ last_p = patch_dict.setdefault(c, None)
+ if last_p == p:
+ continue
+ patch_dict[c] = p
+ if c not in patch_channels:
+ patch_channels.append(c)
+ elif name == "control_change":
+ c, cc, v = event[2:]
+ if not (0 <= c <= 15):
+ continue
+ new_event += [c, cc, v]
+ last_v = control_dict.setdefault((c, cc), 0)
+ if abs(last_v - v) < cc_eps:
+ continue
+ control_dict[(c, cc)] = v
+ elif name == "set_tempo":
+ tempo = event[2]
+ if tempo == 0: # invalid tempo
+ continue
+ bpm = min(int(self.tempo2bpm(tempo)), 383)
+ new_event += [bpm]
+ if abs(last_bpm - bpm) < tempo_eps:
+ continue
+ last_bpm = bpm
+ elif name == "time_signature":
+ nn, dd = event[2:4]
+ if not (1 <= nn <= 16 and 1 <= dd <= 4): # invalid
+ continue
+ nn -= 1 # make it start from 0
+ dd -= 1
+ new_event += [nn, dd]
+ elif name == "key_signature":
+ sf, mi = event[2:]
+ if not (-7 <= sf <= 7 and 0 <= mi <= 1): # invalid
+ continue
+ key_sig_num += 1
+ sf += 7
+ new_event += [sf, mi]
+
+ if name == "note":
+ key = tuple(new_event[:-2])
+ else:
+ key = tuple(new_event[:-1])
+
+ if c != -1:
+ if c not in channels:
+ channels.append(c)
+ tr_map = track_idx_map[c]
+ if track_idx not in tr_map:
+ tr_map[track_idx] = 0
+
+ if event[0] == "note": # to eliminate note overlap due to quantization
+ cp = tuple(new_event[4:6]) # channel pitch
+ if cp in last_notes:
+ last_note_key, last_note = last_notes[cp]
+ last_t = last_note[1] * 16 + last_note[2]
+ last_note[-1] = max(0, min(last_note[-1], t - last_t)) # modify duration
+ if last_note[-1] == 0:
+ event_list.pop(last_note_key)
+ last_notes[cp] = (key, new_event)
+ event_list[key] = new_event
+ event_list = list(event_list.values())
+
+ empty_channels = [c for c in channels if empty_channels[c]]
+
+ if remap_track_channel:
+ patch_channels = []
+ channels_count = 0
+ channels_map = {9: 9} if 9 in channels else {}
+ if remove_empty_channels:
+ channels = sorted(channels, key=lambda x: 1 if x in empty_channels else 0)
+ for c in channels:
+ if c == 9:
+ continue
+ channels_map[c] = channels_count
+ channels_count += 1
+ if channels_count == 9:
+ channels_count = 10
+ channels = list(channels_map.values())
+
+ track_count = 0
+ track_idx_map_order = [k for k, v in sorted(list(channels_map.items()), key=lambda x: x[1])]
+ for c in track_idx_map_order: # tracks not to remove
+ if remove_empty_channels and c in empty_channels:
+ continue
+ tr_map = track_idx_map[c]
+ for track_idx in tr_map:
+ note_tracks = channel_note_tracks[c]
+ if len(note_tracks) != 0 and track_idx not in note_tracks:
+ continue
+ track_count += 1
+ tr_map[track_idx] = track_count
+ for c in track_idx_map_order: # tracks to remove
+ if not (remove_empty_channels and c in empty_channels):
+ continue
+ tr_map = track_idx_map[c]
+ for track_idx in tr_map:
+ note_tracks = channel_note_tracks[c]
+ if not (len(note_tracks) != 0 and track_idx not in note_tracks):
+ continue
+ track_count += 1
+ tr_map[track_idx] = track_count
+
+ empty_channels = [channels_map[c] for c in empty_channels]
+ track_idx_dict = {}
+ key_signature_to_add = []
+ for event in event_list:
+ name = event[0]
+ track_idx = event[3]
+ if name == "note":
+ c = event[4]
+ event[4] = channels_map[c] # channel
+ event[3] = track_idx_map[c][track_idx] # track
+ track_idx_dict.setdefault(event[4], event[3])
+ # setdefault, so the track_idx is first of the channel
+ elif name in ["set_tempo", "time_signature"]:
+ event[3] = 0 # set track 0 for meta events
+ elif name == "key_signature":
+ new_channel_track_idxs = []
+ for c, tr_map in track_idx_map.items():
+ if track_idx in tr_map:
+ new_track_idx = tr_map[track_idx]
+ new_channel_track_idx = (c, new_track_idx)
+ if new_channel_track_idx not in new_channel_track_idxs:
+ new_channel_track_idxs.append(new_channel_track_idx)
+ if len(new_channel_track_idxs) == 0:
+ event[3] = 0
+ continue
+ c, nt = new_channel_track_idxs[0]
+ event[3] = nt
+ if c == 9:
+ event[4] = 7 # sf=0
+ for c, nt in new_channel_track_idxs[1:]:
+ new_event = [*event]
+ new_event[3] = nt
+ if c == 9:
+ new_event[4] = 7 # sf=0
+ key_signature_to_add.append(new_event)
+ elif name == "control_change" or name == "patch_change":
+ c = event[4]
+ event[4] = channels_map[c] # channel
+ tr_map = track_idx_map[c]
+ # move the event to first track of the channel if it's original track is empty
+ note_tracks = channel_note_tracks[c]
+ if len(note_tracks) != 0 and track_idx not in note_tracks:
+ track_idx = channel_note_tracks[c][0]
+ new_track_idx = tr_map.setdefault(track_idx, next(iter(tr_map.values())))
+ event[3] = new_track_idx
+ if name == "patch_change" and event[4] not in patch_channels:
+ patch_channels.append(event[4])
+ event_list += key_signature_to_add
+ track_to_channels ={}
+ for c, tr_map in track_idx_map.items():
+ if c not in channels_map:
+ continue
+ c = channels_map[c]
+ for _, track_idx in tr_map.items():
+ track_to_channels.setdefault(track_idx, [])
+ cs = track_to_channels[track_idx]
+ if c not in cs:
+ cs.append(c)
+
+ if add_default_instr:
+ for c in channels:
+ if c not in patch_channels and c in track_idx_dict:
+ event_list.append(["patch_change", 0, 0, track_idx_dict[c], c, 0])
+
+ if key_sig_num == 0:
+ # detect key signature.
+ root_key = self.detect_key_signature(note_key_hist)
+ if root_key is not None:
+ sf = self.key2sf(root_key, 0)
+ # print("detect_key_signature",sf)
+ for tr, cs in track_to_channels.items():
+ if remap_track_channel and tr == 0:
+ continue
+ event_list.append(["key_signature", 0, 0, tr, (0 if (len(cs) == 1 and cs[0] == 9) else sf) + 7, 0])
+
+ events_name_order = ["time_signature", "key_signature", "set_tempo", "patch_change", "control_change", "note"]
+ events_name_order = {name: i for i, name in enumerate(events_name_order)}
+ events_order = lambda e: e[1:4] + [events_name_order[e[0]]]
+ event_list = sorted(event_list, key=events_order)
+
+ setup_events = {}
+ notes_in_setup = False
+ for i, event in enumerate(event_list): # optimise setup
+ new_event = [*event] # make copy of event
+ if event[0] not in ["note", "time_signature"]:
+ new_event[1] = 0
+ new_event[2] = 0
+ has_next = False
+ has_pre = False
+ if i < len(event_list) - 1:
+ next_event = event_list[i + 1]
+ has_next = event[1] + event[2] == next_event[1] + next_event[2]
+ if notes_in_setup and i > 0:
+ pre_event = event_list[i - 1]
+ has_pre = event[1] + event[2] == pre_event[1] + pre_event[2]
+ if (event[0] == "note" and not has_next) or (notes_in_setup and not has_pre):
+ event_list = sorted(setup_events.values(), key=events_order) + event_list[i:]
+ break
+ else:
+ if event[0] == "note":
+ notes_in_setup = True
+ key = tuple(event[3:-1])
+ setup_events[key] = new_event
+
+ last_t1 = 0
+ midi_seq = []
+ for event in event_list:
+ if remove_empty_channels and event[0] in ["control_change", "patch_change"] and event[4] in empty_channels:
+ continue
+ cur_t1 = event[1]
+ event[1] = event[1] - last_t1
+ tokens = self.event2tokens(event)
+ if not tokens:
+ continue
+ midi_seq.append(tokens)
+ last_t1 = cur_t1
+
+ if add_bos_eos:
+ bos = [self.bos_id] + [self.pad_id] * (self.max_token_seq - 1)
+ eos = [self.eos_id] + [self.pad_id] * (self.max_token_seq - 1)
+ midi_seq = [bos] + midi_seq + [eos]
+ return midi_seq
+
+ def event2tokens(self, event):
+ name = event[0]
+ params = event[1:]
+ if not all([0 <= params[i] < self.event_parameters[p] for i, p in enumerate(self.events[name])]):
+ return []
+ tokens = [self.event_ids[name]] + [self.parameter_ids[p][params[i]]
+ for i, p in enumerate(self.events[name])]
+ tokens += [self.pad_id] * (self.max_token_seq - len(tokens))
+ return tokens
+
+ def tokens2event(self, tokens):
+ if tokens[0] not in self.id_events:
+ return []
+ name = self.id_events[tokens[0]]
+ if len(tokens) <= len(self.events[name]):
+ return []
+ params = tokens[1:]
+ params = [params[i] - self.parameter_ids[p][0] for i, p in enumerate(self.events[name])]
+ if not all([0 <= params[i] < self.event_parameters[p] for i, p in enumerate(self.events[name])]):
+ return []
+ event = [name] + params
+ return event
+
+ def detokenize(self, midi_seq):
+ ticks_per_beat = 480
+ tracks_dict = {}
+ t1 = 0
+ for tokens in midi_seq:
+ if tokens[0] in self.id_events:
+ event = self.tokens2event(tokens)
+ if not event:
+ continue
+ name = event[0]
+ t1 += event[1]
+ t = t1 * 16 + event[2]
+ t = int(t * ticks_per_beat / 16)
+ track_idx = event[3]
+ event_new = [name, t]
+ if name == "note":
+ c, p, v, d = event[4:]
+ d = int(d * ticks_per_beat / 16)
+ event_new += [d, c, p, v]
+ elif name == "control_change" or name == "patch_change":
+ event_new += event[4:]
+ elif name == "set_tempo":
+ event_new += [self.bpm2tempo(event[4])]
+ elif name == "time_signature":
+ nn, dd = event[4:]
+ nn += 1
+ dd += 1
+ event_new += [nn, dd, 24, 8] # usually cc, bb = 24, 8
+ elif name == "key_signature":
+ sf, mi = event[4:]
+ sf -= 7
+ event_new += [sf, mi]
+ else: # should not go here
+ continue
+ if track_idx not in tracks_dict:
+ tracks_dict[track_idx] = []
+ tracks_dict[track_idx].append(event_new)
+ tracks = [tr for idx, tr in sorted(list(tracks_dict.items()), key=lambda it: it[0])]
+
+ for i in range(len(tracks)): # to eliminate note overlap
+ track = tracks[i]
+ track = sorted(track, key=lambda e: e[1])
+ last_note_t = {}
+ zero_len_notes = []
+ for e in reversed(track):
+ if e[0] == "note":
+ t, d, c, p = e[1:5]
+ key = (c, p)
+ if key in last_note_t:
+ d = min(d, max(last_note_t[key] - t, 0))
+ last_note_t[key] = t
+ e[2] = d
+ if d == 0:
+ zero_len_notes.append(e)
+ for e in zero_len_notes:
+ track.remove(e)
+ tracks[i] = track
+ return [ticks_per_beat, *tracks]
+
+ def midi2img(self, midi_score):
+ ticks_per_beat = midi_score[0]
+ notes = []
+ max_time = 1
+ track_num = len(midi_score[1:])
+ for track_idx, track in enumerate(midi_score[1:]):
+ for event in track:
+ t = round(16 * event[1] / ticks_per_beat)
+ if event[0] == "note":
+ d = max(1, round(16 * event[2] / ticks_per_beat))
+ c, p = event[3:5]
+ max_time = max(max_time, t + d + 1)
+ notes.append((track_idx, c, p, t, d))
+ img = np.zeros((128, max_time, 3), dtype=np.uint8)
+ colors = {(i, j): np.random.randint(50, 256, 3) for i in range(track_num) for j in range(16)}
+ for note in notes:
+ tr, c, p, t, d = note
+ img[p, t: t + d] = colors[(tr, c)]
+ img = PIL.Image.fromarray(np.flip(img, 0))
+ return img
+
+ def augment(self, midi_seq, max_pitch_shift=4, max_vel_shift=10, max_cc_val_shift=10, max_bpm_shift=10,
+ max_track_shift=0, max_channel_shift=16):
+ pitch_shift = random.randint(-max_pitch_shift, max_pitch_shift)
+ vel_shift = random.randint(-max_vel_shift, max_vel_shift)
+ cc_val_shift = random.randint(-max_cc_val_shift, max_cc_val_shift)
+ bpm_shift = random.randint(-max_bpm_shift, max_bpm_shift)
+ track_shift = random.randint(0, max_track_shift)
+ channel_shift = random.randint(0, max_channel_shift)
+ midi_seq_new = []
+ key_signature_tokens = []
+ track_to_channels = {}
+ for tokens in midi_seq:
+ tokens_new = [*tokens]
+ if tokens[0] in self.id_events:
+ name = self.id_events[tokens[0]]
+ for i, pn in enumerate(self.events[name]):
+ if pn == "track":
+ tr = tokens[1 + i] - self.parameter_ids[pn][0]
+ tr += track_shift
+ tr = tr % self.event_parameters[pn]
+ tokens_new[1 + i] = self.parameter_ids[pn][tr]
+ elif pn == "channel":
+ c = tokens[1 + i] - self.parameter_ids[pn][0]
+ c0 = c
+ c += channel_shift
+ c = c % self.event_parameters[pn]
+ if c0 == 9:
+ c = 9
+ elif c == 9:
+ c = (9 + channel_shift) % self.event_parameters[pn]
+ tokens_new[1 + i] = self.parameter_ids[pn][c]
+
+ if name == "note":
+ tr = tokens[3] - self.parameter_ids["track"][0]
+ c = tokens[4] - self.parameter_ids["channel"][0]
+ p = tokens[5] - self.parameter_ids["pitch"][0]
+ v = tokens[6] - self.parameter_ids["velocity"][0]
+ if c != 9: # no shift for drums
+ p += pitch_shift
+ if not 0 <= p < 128:
+ return midi_seq
+ v += vel_shift
+ v = max(1, min(127, v))
+ tokens_new[5] = self.parameter_ids["pitch"][p]
+ tokens_new[6] = self.parameter_ids["velocity"][v]
+ track_to_channels.setdefault(tr, [])
+ cs = track_to_channels[tr]
+ if c not in cs:
+ cs.append(c)
+ elif name == "control_change":
+ cc = tokens[5] - self.parameter_ids["controller"][0]
+ val = tokens[6] - self.parameter_ids["value"][0]
+ if cc in [1, 2, 7, 11]:
+ val += cc_val_shift
+ val = max(1, min(127, val))
+ tokens_new[6] = self.parameter_ids["value"][val]
+ elif name == "set_tempo":
+ bpm = tokens[4] - self.parameter_ids["bpm"][0]
+ bpm += bpm_shift
+ bpm = max(1, min(383, bpm))
+ tokens_new[4] = self.parameter_ids["bpm"][bpm]
+ elif name == "key_signature":
+ sf = tokens[4] - self.parameter_ids["sf"][0]
+ mi = tokens[5] - self.parameter_ids["mi"][0]
+ sf -= 7
+ k = self.sf2key(sf)
+ k = (k + pitch_shift) % 12
+ sf = self.key2sf(k, mi)
+ sf += 7
+ tokens_new[4] = self.parameter_ids["sf"][sf]
+ tokens_new[5] = self.parameter_ids["mi"][mi]
+ key_signature_tokens.append(tokens_new)
+ midi_seq_new.append(tokens_new)
+ for tokens in key_signature_tokens:
+ tr = tokens[3] - self.parameter_ids["track"][0]
+ if tr in track_to_channels:
+ cs = track_to_channels[tr]
+ if len(cs) == 1 and cs[0] == 9:
+ tokens[4] = self.parameter_ids["sf"][7] # sf=0
+ return midi_seq_new
+
+ def check_quality(self, midi_seq, alignment_min=0.3, tonality_min=0.8, piano_max=0.7, notes_bandwidth_min=3,
+ notes_density_max=50, notes_density_min=2.5, total_notes_max=20000, total_notes_min=256,
+ note_window_size=16):
+ total_notes = 0
+ channels = []
+ time_hist = [0] * 16
+ note_windows = {}
+ notes_sametime = []
+ notes_density_list = []
+ tonality_list = []
+ notes_bandwidth_list = []
+ instruments = {}
+ piano_channels = []
+ abs_t1 = 0
+ last_t = 0
+ for tsi, tokens in enumerate(midi_seq):
+ event = self.tokens2event(tokens)
+ if not event:
+ continue
+ t1, t2, tr = event[1:4]
+ abs_t1 += t1
+ t = abs_t1 * 16 + t2
+ c = None
+ if event[0] == "note":
+ c, p, v, d = event[4:]
+ total_notes += 1
+ time_hist[t2] += 1
+ if c != 9: # ignore drum channel
+ if c not in instruments:
+ instruments[c] = 0
+ if c not in piano_channels:
+ piano_channels.append(c)
+ note_windows.setdefault(abs_t1 // note_window_size, []).append(p)
+ if last_t != t:
+ notes_sametime = [(et, p_) for et, p_ in notes_sametime if et > last_t]
+ notes_sametime_p = [p_ for _, p_ in notes_sametime]
+ if len(notes_sametime) > 0:
+ notes_bandwidth_list.append(max(notes_sametime_p) - min(notes_sametime_p))
+ notes_sametime.append((t + d - 1, p))
+ elif event[0] == "patch_change":
+ c, p = event[4:]
+ instruments[c] = p
+ if p == 0 and c not in piano_channels:
+ piano_channels.append(c)
+ if c is not None and c not in channels:
+ channels.append(c)
+ last_t = t
+ reasons = []
+ if total_notes < total_notes_min:
+ reasons.append("total_min")
+ if total_notes > total_notes_max:
+ reasons.append("total_max")
+ if len(note_windows) == 0 and total_notes > 0:
+ reasons.append("drum_only")
+ if reasons:
+ return False, reasons
+ time_hist = sorted(time_hist, reverse=True)
+ alignment = sum(time_hist[:2]) / total_notes
+ for notes in note_windows.values():
+ key_hist = [0] * 12
+ for p in notes:
+ key_hist[p % 12] += 1
+ key_hist = sorted(key_hist, reverse=True)
+ tonality_list.append(sum(key_hist[:7]) / len(notes))
+ notes_density_list.append(len(notes) / note_window_size)
+ tonality_list = sorted(tonality_list)
+ tonality = sum(tonality_list) / len(tonality_list)
+ notes_bandwidth = sum(notes_bandwidth_list) / len(notes_bandwidth_list) if notes_bandwidth_list else 0
  notes_density = max(notes_density_list) if notes_density_list else 0
  piano_ratio = len(piano_channels) / len(channels)
- if len(channels) <=3: # ignore piano threshold if it is a piano solo midi
+ if len(channels) <= 3: # ignore piano threshold if it is a piano solo midi
  piano_max = 1
- if alignment < alignment_min: # check weather the notes align to the bars (because some midi files are recorded)
+ if alignment < alignment_min: # check weather the notes align to the bars (because some midi files are recorded)
  reasons.append("alignment")
  if tonality < tonality_min: # check whether the music is tonal
  reasons.append("tonality")
@@ -464,6 +1116,16 @@ class MIDITokenizer:
  reasons.append("bandwidth")
  if not notes_density_min < notes_density < notes_density_max:
  reasons.append("density")
- if piano_ratio > piano_max: # check whether most instruments is piano (because some midi files don't have instruments assigned correctly)
+ if piano_ratio > piano_max: # check whether most instruments is piano (because some midi files don't have instruments assigned correctly)
  reasons.append("piano")
  return not reasons, reasons
+
+
+class MIDITokenizer:
+ def __new__(cls, version="v2"):
+ if version == "v1":
+ return MIDITokenizerV1()
+ elif version == "v2":
+ return MIDITokenizerV2()
+ else:
+ raise ValueError(f"Unsupported version: {version}")