File size: 1,925 Bytes
f1cc9cf | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 | import pandas as pd
import torchaudio
from glob import glob
import os
import csv
from tqdm import tqdm
import pandas
sentence_bounds = {'!', '.', ';', '?', '…'}
files = list(sorted(glob("earnings22/media/*.mp3")))
metadata = []
for audio_file in tqdm(files):
file_id = audio_file.split("/")[-1].split(".")[0]
nlp_file = f"earnings22/aligned/{file_id}.nlp"
speech, sr = torchaudio.load(audio_file)
with open(nlp_file, "r") as nlp:
start, end = None, None
sentence = ""
segment_id = 0
csvreader = csv.DictReader(nlp, delimiter="|")
for row in csvreader:
punct = row["punctuation"].strip()
sentence += row["token"]
if punct:
sentence += punct
sentence += " "
if start is None and row["ts"].strip():
start = float(row["ts"]) - 0.1
if row["endTs"].strip():
end = float(row["endTs"]) + 0.1
if punct in sentence_bounds and start is not None and end is not None:
segment = speech[:, int(start*sr):int(end*sr)+1].contiguous()
os.makedirs(f"earnings22/segmented/{file_id}", exist_ok=True)
torchaudio.save(f"earnings22/segmented/{file_id}/{segment_id}.wav",
segment, sr, encoding="PCM_S", bits_per_sample=16)
data_row = {
"source_id": f"{file_id}",
"segment_id": segment_id,
"file": f"{file_id}/{segment_id}.wav",
"start_ts": start,
"end_ts": end,
"sentence": sentence.strip()
}
metadata.append(data_row)
start, end = None, None
sentence = ""
segment_id += 1
pd.DataFrame(metadata).to_csv("earnings22/segmented/metadata.csv", index=False)
|