utils.py

import tempfile
import collections

import librosa

import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.patches import Rectangle
from PIL import Image

class AudioIOReadError(BaseException):  # pylint:disable=g-bad-exception-name
  pass

def upload_audio(audio, sample_rate):
  
  return wav_data_to_samples_librosa(audio, sample_rate=sample_rate)
  
def wav_data_to_samples_librosa(audio_file, sample_rate):
  """Loads an in-memory audio file with librosa.
  Use this instead of wav_data_to_samples if the wav is 24-bit, as that's
  incompatible with wav_data_to_samples internal scipy call.
  Will copy to a local temp file before loading so that librosa can read a file
  path. Librosa does not currently read in-memory files.
  It will be treated as a .wav file.
  Args:
    audio_file: Wav file to load.
    sample_rate: The number of samples per second at which the audio will be
        returned. Resampling will be performed if necessary.
  Returns:
    A numpy array of audio samples, single-channel (mono) and sampled at the
    specified rate, in float32 format.
  Raises:
    AudioIOReadException: If librosa is unable to load the audio data.
  """
  with tempfile.NamedTemporaryFile(suffix='.wav') as wav_input_file:
    wav_input_file.write(audio_file)
    # Before copying the file, flush any contents
    wav_input_file.flush()
    # And back the file position to top (not need for Copy but for certainty)
    wav_input_file.seek(0)
    return load_audio(wav_input_file.name, sample_rate)

def load_audio(audio_filename, sample_rate, duration=10):
  """Loads an audio file.
  Args:
    audio_filename: File path to load.
    sample_rate: The number of samples per second at which the audio will be
        returned. Resampling will be performed if necessary.
  Returns:
    A numpy array of audio samples, single-channel (mono) and sampled at the
    specified rate, in float32 format.
  Raises:
    AudioIOReadError: If librosa is unable to load the audio data.
  """
  try:
    y, unused_sr = librosa.load(audio_filename, sr=sample_rate, mono=True, duration=duration)
  except Exception as e:  # pylint: disable=broad-except
    raise AudioIOReadError(e)
  return y

# Generate piano_roll
def sequence_to_pandas_dataframe(sequence):
    pd_dict = collections.defaultdict(list)
    for note in sequence.notes:
        pd_dict["start_time"].append(note.start_time)
        pd_dict["end_time"].append(note.end_time)
        pd_dict["duration"].append(note.end_time - note.start_time)
        pd_dict["pitch"].append(note.pitch)
        
    return pd.DataFrame(pd_dict)

def dataframe_to_pianoroll_img(df):
    fig = plt.figure(figsize=(8, 5))
    ax = fig.add_subplot(111)
    ax.scatter(df.start_time, df.pitch, c="white")
    for _, row in df.iterrows():
        ax.add_patch(Rectangle((row["start_time"], row["pitch"]-0.4), row["duration"], 0.4, color="black"))
    plt.xlabel('time (sec.)', fontsize=18)
    plt.ylabel('pitch (MIDI)', fontsize=16)
    return fig

def fig2img(fig):
    """Convert a Matplotlib figure to a PIL Image and return it"""
    import io
    buf = io.BytesIO()
    fig.savefig(buf, format="png")
    buf.seek(0)
    img = Image.open(buf)
    return img

def create_image_from_note_sequence(sequence):
    df_sequence = sequence_to_pandas_dataframe(sequence)
    fig = dataframe_to_pianoroll_img(df_sequence)
    img = fig2img(fig)
    return img