rssdb / test /upload_to_hf.py
devfed's picture
Add files using upload-large-folder tool
d27919e verified
#content for diph1.txt
# 001. Nu este treaba lor ce constituţie avem.
# 002. Ea era tot timpul pe minge.
# ...
# 499. Ea e singurul meu sprijin, fără ea eram ţărână.
# 500. A, de fapt ar fi ceva.
#001 is the audio_index
#example paths for each collection:
# TEXT_FOLDER + '/' + COLLECTION[0] + '.txt' (from here we take audio_index )
# ACCENT_FOLDER + '/' + COLLECTION[0] + '/adr_' + COLLECTION[0] + '_' + audio_index
#...
# AUDIO_FOLDER
#Lets build a metadata.csv with column:
#file_name,duration,start_time,end_time,text,accent,phonemes,hts_labels,hts_label_path
#audio_001,23.35,0.0,23.35,nu jeste..., ,n u j e...,1500000 2200000 n | ...,audio_001.lab
import os
import csv
# Constants
ACCENT_FOLDER = 'accent'
AUDIO_FOLDER = 'audio'
SYN_AUDIO_FOLDER = 'synthesized_audio'
HTS_LABELS_FOLDER = 'hts_labels'
PHONEMES_FOLDER = 'phoneme'
TEXT_FOLDER = 'text'
COLLECTIONS = ['news', 'novel', 'SUS']
OUTPUT_CSV = 'metadata.csv'
def read_text_file(text_path):
"""Returns a dict: audio_index -> sentence"""
mapping = {}
with open(text_path, 'r', encoding='utf-8') as f:
for line in f:
if '. ' in line:
idx, sentence = line.strip().split('. ', 1)
idx = idx.zfill(3)
mapping[idx] = sentence.strip()
return mapping
def read_phoneme_file(filepath):
phonemes = []
with open(filepath, 'r', encoding='utf-8') as f:
for line in f:
parts = line.strip().split()
if len(parts) == 3:
start, end, label = parts
start_us = int(start)
end_us = int(end)
start_sec = start_us / 1e6
end_sec = end_us / 1e6
if label != "#":
phonemes.append((start_sec, end_sec, label))
return phonemes
def main():
rows = []
for collection in COLLECTIONS:
print(f"Processing collection: {collection}")
text_map = read_text_file(f"{TEXT_FOLDER}/{collection}.txt")
for audio_index, text in text_map.items():
file_name = f"{collection}_{audio_index}"
# Paths
phoneme_path = os.path.join(PHONEMES_FOLDER, collection, f"adr_{collection}_{audio_index}.phs")
accent_path = os.path.join(ACCENT_FOLDER, collection, f"adr_{collection}_{audio_index}")
hts_label_path = os.path.join(HTS_LABELS_FOLDER, collection, f"adr_{file_name}.lab") # Path calculated just for reference
accent = ""
if os.path.exists(accent_path):
with open(accent_path, 'r', encoding='utf-8') as f:
accent = f.read().strip()
rows.append([
AUDIO_FOLDER + '/' + collection + '/adr_' + file_name + '.wav',
text,
accent,
phoneme_path,
hts_label_path
])
# Write metadata CSV
with open(OUTPUT_CSV, 'w', newline='', encoding='utf-8') as f:
writer = csv.writer(f, quoting=csv.QUOTE_ALL)
writer.writerow(["file_name", "text", "accent", "phonemes", "hts_label_path"])
for row in rows:
writer.writerow(row)
print(f"\n✅ metadata.csv written with {len(rows)} entries.")
if __name__ == "__main__":
main()