| import os |
| import random |
| import pdfplumber |
| from datasets import load_dataset |
|
|
| BASE_DIR = "/home/cloud/StyleTTS2-fine-tuning" |
| DATA_DIR = os.path.join(BASE_DIR, "Data") |
| INPUT_PDF = os.path.join(DATA_DIR, "English_CORE2000.pdf") |
| FINAL_OUTPUT_FILE = os.path.join(DATA_DIR, "source_text_final.txt") |
| HF_DATASET_ID = "agentlans/high-quality-english-sentences" |
| HF_DOWNLOAD_COUNT = 8000 |
|
|
| os.makedirs(DATA_DIR, exist_ok=True) |
|
|
| def extract_pdf_sentences(): |
| pdf_sentences = [] |
| |
| if not os.path.exists(INPUT_PDF): |
| print(f"Warning: PDF file not found at {INPUT_PDF}. Skipping.") |
| return [] |
|
|
| with pdfplumber.open(INPUT_PDF) as pdf: |
| for page in pdf.pages: |
| words = page.extract_words() |
| header_x = None |
| header_bottom = None |
| |
| for j, word in enumerate(words): |
| if word['text'] == 'Sample' and j+1 < len(words): |
| next_word = words[j+1] |
| if next_word['text'] == 'Sentence': |
| header_x = word['x0'] |
| header_bottom = word['bottom'] |
| break |
| |
| if header_x is not None: |
| crop_box = (header_x - 5, header_bottom + 5, page.width, page.height) |
| try: |
| cropped_page = page.crop(crop_box) |
| text_block = cropped_page.extract_text() |
| if text_block: |
| lines = text_block.split('\n') |
| for line in lines: |
| clean_line = line.strip() |
| if len(clean_line) > 10: |
| pdf_sentences.append(clean_line) |
| except ValueError: |
| pass |
| |
| print(f"Extracted {len(pdf_sentences)} raw lines from PDF.") |
| return pdf_sentences |
|
|
| def clean_sentences(sentences): |
| print("Cleaning") |
| cleaned_list = [] |
| removed_count = 0 |
| |
| for line in sentences: |
| cleaned = line.replace('EnglishClass101.com', '') |
| stripped = cleaned.strip() |
| |
| if not stripped: |
| continue |
| |
| if len(stripped.split()) < 4: |
| removed_count += 1 |
| continue |
| |
| cleaned_list.append(stripped) |
| |
| print(f"Removed {removed_count} short sentences. Kept {len(cleaned_list)}.") |
| return cleaned_list |
|
|
| def get_hf_sentences(count): |
| print(f"Downloading {count} lines from Hugging Face") |
| hf_sentences = [] |
| |
| try: |
| ds = load_dataset(HF_DATASET_ID, split=f"train[:{count}]") |
| text_column = "text" |
| if "sentence" in ds.column_names: |
| text_column = "sentence" |
| elif "content" in ds.column_names: |
| text_column = "content" |
|
|
| for row in ds: |
| line = row[text_column] |
| if line and isinstance(line, str): |
| clean_line = line.strip() |
| if clean_line: |
| hf_sentences.append(clean_line) |
| |
| print(f"Downloaded {len(hf_sentences)} lines from HF.") |
| except Exception as e: |
| print(f"Error downloading HF dataset: {e}") |
| |
| return hf_sentences |
|
|
| def merge_and_save(): |
| raw_pdf_lines = extract_pdf_sentences() |
| clean_pdf_lines = clean_sentences(raw_pdf_lines) |
| hf_lines = get_hf_sentences(HF_DOWNLOAD_COUNT) |
| |
| print("Merging and Shuffling ") |
| combined_data = clean_pdf_lines + hf_lines |
| random.shuffle(combined_data) |
| |
| print(f"Saving {len(combined_data)} total lines to {FINAL_OUTPUT_FILE}...") |
| with open(FINAL_OUTPUT_FILE, "w", encoding="utf-8") as f: |
| f.write("\n".join(combined_data)) |
| |
| print("Dataset generation complete.") |
|
|
| if __name__ == "__main__": |
| merge_and_save() |
