ravdess / ravdess_preprocessor.py
narad's picture
Upload ravdess_preprocessor.py
5a839bf
raw
history blame
9.21 kB
# from pathlib import Path
# import pandas as pd
# import regex as re
# import os
# import torchaudio
# import argparse
# from tqdm import tqdm
# from collections import OrderedDict
# feat_dict = OrderedDict()
# od['Modality'] = ['full-AV', 'video-only', 'audio-only']
# od['Vocal channel'] = ['speech', 'song']
# od['Emotion'] = ['neutral', 'calm', 'happy', 'sad', 'angry', 'fearful', 'disgust', 'surprised']
# od['Emotion intensity'] = ['normal', 'strong']
# od['Statement'] = ["Kids are talking by the door", "Dogs are sitting by the door"]
# od['Repetition'] = ["1st repetition", "2nd repetition"]
# # def filename2feats(filename):
# # codes = filename.stem.split('-')
# # for i, k in enumerate(od.keys()):
# # d = {}
# # d[k] = od[k][int(codes[i])-1]
# # d['Actor'] = codes[-1]
# # d['Gender'] = 'female' if int(codes[-1]) % 2 == 0 else 'male'
# # return d
# def preprocess(data_root_path):
# output_dir = data_root_path / "RAVDESS_ser"
# for f in data_root_path.iterdir():
# print(f)
# filename2feats(filename)
# print("\n\n")
# # Filename identifiers
# # Modality (01 = full-AV, 02 = video-only, 03 = audio-only).
# # Vocal channel (01 = speech, 02 = song).
# # Emotion (01 = neutral, 02 = calm, 03 = happy, 04 = sad, 05 = angry, 06 = fearful, 07 = disgust, 08 = surprised).
# # Emotional intensity (01 = normal, 02 = strong). NOTE: There is no strong intensity for the 'neutral' emotion.
# # Statement (01 = "Kids are talking by the door", 02 = "Dogs are sitting by the door").
# # Repetition (01 = 1st repetition, 02 = 2nd repetition).
# # Actor (01 to 24. Odd numbered actors are male, even numbered actors are female).
# # Filename example: 02-01-06-01-02-01-12.mp4
# # Video-only (02)
# # Speech (01)
# # Fearful (06)
# # Normal intensity (01)
# # Statement "dogs" (02)
# # 1st Repetition (01)
# # 12th Actor (12)
# # Female, as the actor ID number is even.
# # self.data_root_path = Path(data_root_path)
# # df = pd.DataFrame()
# # for session in range(1,5):
# # print(f"Processing session {session}")
# # df = pd.concat([df, self.read_session_data(session)])
# # # Write the sliced wavs
# # print("Writing wav slices to file...")
# # sample_rate = 16000
# # for index, row in df.iterrows():
# # old_filename = str(self.data_root_path / Path(row['Path_to_Wav']))
# # new_filename = str(output_dir / (index + ".wav"))
# # waveform = self.read_audio(old_filename,
# # start=row['Time_Start'],
# # end=row['Time_End'])
# # torchaudio.save(os.path.join(new_filename),
# # src=waveform,
# # sample_rate=sample_rate)
# # df.at[index, 'Path_to_Wav'] = new_filename
# # # Write out the combined data information
# # try:
# # df.to_csv(output_filename, index=False, header=True)
# # except:
# # print("Error writing dataframe to csv.")
# # def read_session_data(self, session_id):
# # d1 = self.read_emotion_labels(session_id)
# # d2 = self.read_transcriptions(session_id)
# # return d1.join(d2)
# # def read_emotion_labels(self, session_id):
# # emo_path = Path(self.data_root_path / Path(f"Session{session_id}") / Path("dialog") / Path("EmoEvaluation"))
# # emo_files = [f for f in list(emo_path.iterdir()) if f.suffix == ".txt"]
# # df = pd.DataFrame()
# # for ef in emo_files:
# # df2 = self.read_emotion_file(ef)
# # for ri, row in df2.iterrows():
# # df2.loc[ri, 'Path_to_Wav'] = os.path.join(f"Session{session_id}",
# # "dialog", "wav",
# # row['Session_ID'] +".wav")
# # df = pd.concat([df, df2])
# # df = df.set_index('ID')
# # return df
# # def slice_audio(self, session_id):
# # for i, row in df.iterrows():
# # filename = row['Session_ID'] + ".wav"
# # wav_path = Path(self.data_root_path / Path(f"Session{session_id}") / Path("dialog") / Path("wav") / Path(filename))
# # print("wav path = ", wav_path)
# # self.read_audio(wav_path, row['Time_Start'], row['Time_End'], row['Annotations'])
# # def read_emotion_file(self, filename):
# # time_extract_pattern = "\[([0-9\.]+) - ([0-9\.]+)\] +([^ ]+) +([^ ]+) \[([^\]]+)\]"
# # df = pd.DataFrame() #columns=columns)
# # i = 0
# # with open(filename) as file:
# # lines = file.readlines()
# # lines = lines[2:] #:10]
# # while i < len(lines):
# # # Remove header
# # if match := re.search(time_extract_pattern, lines[i].replace("\t", " ")):
# # time_start = float(match.group(1))
# # time_end = float(match.group(2))
# # filename = match.group(3)
# # mys_id = match.group(4)
# # digits = [float(x) for x in match.group(5).split(", ")]
# # annotations = []
# # while lines[i] != "\n":
# # i += 1
# # if lines[i].startswith("C-"):
# # aid, anns, _ = lines[i].split("\t")
# # for an in anns.split(";")[:-1]:
# # annotations.append(an.strip())
# # elif lines[i].startswith("A-"):
# # pass
# # annotations = list(set(annotations))
# # annotations = ','.join(annotations)
# # session_id = filename[:filename.rindex("_")]
# # utt_id = filename[filename.rindex("_")+1:]
# # df2 = pd.DataFrame([{
# # 'ID': filename, # ID for join between dataframes is the filename
# # 'Session_ID': session_id,
# # 'Utterance_ID': utt_id,
# # 'Time_Start': time_start,
# # 'Time_End': time_end,
# # 'Labels': annotations}])
# # df = pd.concat([df, df2], ignore_index=True)
# # else:
# # i += 1
# # return df
# # def read_transcriptions(self, session_id):
# # df = pd.DataFrame()
# # transcripts_path = Path(self.data_root_path / Path(f"Session{session_id}") / Path("dialog") / Path("transcriptions"))
# # transcript_files = [f for f in list(transcripts_path.iterdir()) if f.suffix == ".txt"]
# # for f in transcript_files:
# # df = pd.concat([df, self.read_transcript(f)], ignore_index=True)
# # df = df.set_index('ID')
# # return df
# # def read_transcript(self, filename):
# # df = pd.DataFrame()
# # with open(filename, "r") as f:
# # for l in f.readlines():
# # cols = l.strip().split(" ")
# # if l[1] != ":" and len(cols) > 2: # There are some lines like "F:Mmhmm." that get ignored here
# # df2 = pd.DataFrame([{
# # 'ID': cols[0],
# # 'Transcription': ' '.join(cols[2:])
# # }])
# # df = pd.concat([df, df2])
# # return df
# # def read_audio(self, filename, start, end, sample_rate=16000):
# # waveform, sample_rate = torchaudio.load(filename,
# # frame_offset=int(start * sample_rate),
# # num_frames=int((end-start) * sample_rate))
# # return waveform
# # if __name__ == '__main__':
# # # osx_path = '/Users/narad/Downloads/RAVDESS_full_release'
# # # windows_path = r'C:\Users\jasonn\Desktop\ser\data\RAVDESS_full_release'
# # parser = argparse.ArgumentParser(description='Process some integers.')
# # parser.add_argument('--data_dir', type=Path, required=True,
# # help='Path to IEOMCAP release directory.')
# # parser.add_argument('--output_file', type=Path, default="data.csv",
# # help='Filename for Huggingface-compatible dataset csv file.')
# # parser.add_argument('--output_dir', type=Path, default="processed",
# # help='Directory for processed wav files')
# # args = parser.parse_args()
# # print(args)
# # reader = RAVDESS(data_root_path=args.data_dir,
# # output_filename=args.output_file)
# # columns = ['Utterance_ID',
# # 'Time_Start',
# # 'Time-End',
# # 'Annotations']