cantomap / convert.py
wcyat's picture
initial commit
b586e31 unverified
import xml.etree.ElementTree as ET
import csv
import pandas as pd
from pydub import AudioSegment
import re
import sys
from typing import List, Tuple, Dict
# Global dictionary to store custom character choices for each pinyin
custom_choices: Dict[str, str] = {}
def parse_xml(xml_content: str) -> List[Tuple[int, int, str]]:
root = ET.fromstring(xml_content)
time_order = {ts.get('TIME_SLOT_ID'): int(ts.get('TIME_VALUE'))
for ts in root.iter('TIME_SLOT')}
tier_data = []
for tier in root.iter('TIER'):
tier_id = tier.get('TIER_ID')
if not ('-' in tier_id) and (tier_id == 'G' or tier_id == 'E'):
for annotation in tier.iter('ALIGNABLE_ANNOTATION'):
start_time = time_order[annotation.get('TIME_SLOT_REF1')]
end_time = time_order[annotation.get('TIME_SLOT_REF2')]
text = annotation.find('./ANNOTATION_VALUE').text
tier_data.append((start_time, end_time, text))
return tier_data
def transform_latin_to_chinese(text: str, dict_df: pd.DataFrame, output_wav: str) -> Tuple[str, pd.DataFrame]:
global custom_choices
transformed_text = ""
i = 0
while i < len(text):
char = text[i]
if char == "&":
j = i + 1
while j < i + 7 and j < len(text) and text[j].isalpha():
j += 1
if j < len(text) and text[j].isdigit():
term = text[(i+1):j + 1].lower()
pinyin_entries = dict_df[dict_df['拼音'] == term]['繁體']
if not pinyin_entries.empty:
if len(pinyin_entries) > 1:
full_sentence = f"Full Sentence: {text}"
print(full_sentence)
print(f"Path: {output_wav}")
print(f"Multiple entries found for {
term}. Choose one (or enter a custom character):")
for idx, entry in enumerate(pinyin_entries):
print(f"{idx + 1}. {entry}")
print(f"{len(pinyin_entries) +
1}. Enter a custom character")
choice = input("Enter the number of your choice: ")
if choice.isdigit() and int(choice) <= len(pinyin_entries):
choice = int(choice) - 1
transformed_text += pinyin_entries.values[choice]
custom_choices[term] = pinyin_entries.values[choice]
elif choice == str(len(pinyin_entries) + 1):
custom_choice = input(
f"Enter a custom character for {term}: ")
transformed_text += custom_choice
custom_choices[term] = custom_choice
# Update DataFrame with the custom character
dict_df = pd.concat([dict_df, pd.DataFrame(
[{'拼音': term, '繁體': custom_choice}])], ignore_index=True)
else:
print("Invalid choice. Using the default character.")
transformed_text += pinyin_entries.values[0]
else:
try:
transformed_text += pinyin_entries.values[0] or ""
except:
pass
else:
print(f"Full Sentence: {text}")
print(f"Path: {output_wav}")
custom_choice = input(f"No corresponding character found for {
term}. Enter a custom character: ")
transformed_text += custom_choice
custom_choices[term] = custom_choice
# Update DataFrame with the custom character
dict_df = pd.concat([dict_df, pd.DataFrame(
[{'拼音': term, '繁體': custom_choice}])], ignore_index=True)
i = j + 1
else:
transformed_text += char
i += 1
else:
transformed_text += char
i += 1
return transformed_text, dict_df
def clean_transformed_text(text: str) -> str:
# Replace "#" with a blank space
text = text.replace("#", " ")
# Remove special characters and symbols
text = re.sub(r"[^a-zA-Z0-9\u4e00-\u9fff ]", "", text)
# Remove HTML entities (e.g., &amp;)
text = re.sub(r"&[a-zA-Z]+;", "", text)
return text
def extract_audio_segments(input_wav: str, output_wav: str, start_time: int, end_time: int) -> None:
audio = AudioSegment.from_wav(input_wav)
segment = audio[start_time:end_time]
segment.export(output_wav, format="wav")
def main() -> None:
with open(sys.argv[1], "r", encoding="utf-8") as file:
xml_content = file.read()
tier_data = parse_xml(xml_content)
with open("./粵語字典_(耶魯_數字).csv", "r", encoding="utf-8") as dict_file:
dict_csv = dict_file.name
dict_df = pd.read_csv(dict_csv)
# Replace with your actual audio file
audio_file = sys.argv[1].replace("eaf", "wav")
transformed_tier_data = []
for i, (start_time, end_time, text) in enumerate(tier_data):
output_wav = f"audio/{sys.argv[1].split('/')
[-1].replace('.eaf', '')}_ts{i+1}.wav"
extract_audio_segments(audio_file, output_wav, start_time, end_time)
transformed_text, dict_df = transform_latin_to_chinese(text, dict_df, output_wav)
transformed_text = clean_transformed_text(transformed_text)
transformed_tier_data.append(
(start_time, end_time, transformed_text, output_wav.split("/")[-1]))
# Save the updated DataFrame to the CSV file
dict_df.to_csv(dict_csv, index=False, encoding='utf-8')
tsv_filename = "transcript/" + \
sys.argv[1].split("/")[-1].replace("eaf", "tsv")
with open(tsv_filename, "w", newline="", encoding="utf-8") as tsvfile:
tsv_writer = csv.writer(tsvfile, delimiter='\t')
tsv_writer.writerow(
["timestamp_start", "timestamp_end", "text", "path"])
tsv_writer.writerows(transformed_tier_data)
if __name__ == "__main__":
main()