app.py

import os
os.environ["NUMBA_DISABLE_CACHE"] = "1"

import os
import gradio as gr
from docx import Document
from TTS.api import TTS
import tempfile
import zipfile
from io import BytesIO
import re

# Voice model
VOICE_MODEL = "tts_models/en/vctk/vits"

# Embedded metadata (from your file)
SPEAKER_METADATA = {
	    300: { "age": 23, "gender": "F", "accent": "American"},
	    271: { "age": 19, "gender": "M", "accent": "Scottish"},
	    287: { "age": 23, "gender": "M", "accent": "English"},
	    262: { "age": 23, "gender": "F", "accent": "Scottish"},
	    284: { "age": 20, "gender": "M", "accent": "Scottish"},
	    297: { "age": 20, "gender": "F", "accent": "American"},
	    227: { "age": 38, "gender": "M", "accent": "English"},
	    246: { "age": 22, "gender": "M", "accent": "Scottish"},
	    225: { "age": 23, "gender": "F", "accent": "English"},
	    259: { "age": 23, "gender": "M", "accent": "English"},
	    252: { "age": 22, "gender": "M", "accent": "Scottish"},
	    231: { "age": 23, "gender": "F", "accent": "English"},
	    266: { "age": 22, "gender": "F", "accent": "Irish"},
	    241: { "age": 21, "gender": "M", "accent": "Scottish"},
	    312: { "age": 19, "gender": "F", "accent": "Canadian"},
	    329: { "age": 23, "gender": "F", "accent": "American"},
	    232: { "age": 23, "gender": "M", "accent": "English"},
	    305: { "age": 19, "gender": "F", "accent": "American"},
	    311: { "age": 21, "gender": "M", "accent": "American"},
	    301: { "age": 23, "gender": "F", "accent": "American"},
	    304: { "age": 22, "gender": "M", "accent": "NorthernIrish"},
	    310: { "age": 21, "gender": "F", "accent": "American"},
	    260: { "age": 21, "gender": "M", "accent": "Scottish"},
	    315: { "age": 18, "gender": "M", "accent": "American"},
	    374: { "age": 28, "gender": "M", "accent": "Australian"},
	    364: { "age": 23, "gender": "M", "accent": "Irish"},
	    269: { "age": 20, "gender": "F", "accent": "English"},
	    345: { "age": 22, "gender": "M", "accent": "American"},
	    326: { "age": 26, "gender": "M", "accent": "Australian"},
	    343: { "age": 27, "gender": "F", "accent": "Canadian"},
	    230: { "age": 22, "gender": "F", "accent": "English"},
	    376: { "age": 22, "gender": "M", "accent": "Indian"},
	    240: { "age": 21, "gender": "F", "accent": "English"},
	    298: { "age": 19, "gender": "M", "accent": "Irish"},
	    272: { "age": 23, "gender": "M", "accent": "Scottish"},
	    248: { "age": 23, "gender": "F", "accent": "Indian"},
	    264: { "age": 23, "gender": "F", "accent": "Scottish"},
	    250: { "age": 22, "gender": "F", "accent": "English"},
	    292: { "age": 23, "gender": "M", "accent": "NorthernIrish"},
	    237: { "age": 22, "gender": "M", "accent": "Scottish"},
	    363: { "age": 22, "gender": "M", "accent": "Canadian"},
	    313: { "age": 24, "gender": "F", "accent": "Irish"},
	    285: { "age": 21, "gender": "M", "accent": "Scottish"},
	    268: { "age": 23, "gender": "F", "accent": "English"},
	    302: { "age": 20, "gender": "M", "accent": "Canadian"},
	    261: { "age": 26, "gender": "F", "accent": "NorthernIrish"},
	    336: { "age": 18, "gender": "F", "accent": "SouthAfrican"},
	    288: { "age": 22, "gender": "F", "accent": "Irish"},
	    226: { "age": 22, "gender": "M", "accent": "English"},
	    277: { "age": 23, "gender": "F", "accent": "English"},
	    360: { "age": 19, "gender": "M", "accent": "American"},
	    257: { "age": 24, "gender": "F", "accent": "English"},
	    254: { "age": 21, "gender": "M", "accent": "English"},
	    339: { "age": 21, "gender": "F", "accent": "American"},
	    323: { "age": 19, "gender": "F", "accent": "SouthAfrican"},
	    255: { "age": 19, "gender": "M", "accent": "Scottish"},
	    249: { "age": 22, "gender": "F", "accent": "Scottish"},
	    293: { "age": 22, "gender": "F", "accent": "NorthernIrish"},
	    244: { "age": 22, "gender": "F", "accent": "English"},
	    245: { "age": 25, "gender": "M", "accent": "Irish"},
	    361: { "age": 19, "gender": "F", "accent": "American"},
	    314: { "age": 26, "gender": "F", "accent": "SouthAfrican"},
	    308: { "age": 18, "gender": "F", "accent": "American"},
	    229: { "age": 23, "gender": "F", "accent": "English"},
	    341: { "age": 26, "gender": "F", "accent": "American"},
	    275: { "age": 23, "gender": "M", "accent": "Scottish"},
	    263: { "age": 22, "gender": "M", "accent": "Scottish"},
	    253: { "age": 22, "gender": "F", "accent": "Welsh"},
	    299: { "age": 25, "gender": "F", "accent": "American"},
	    316: { "age": 20, "gender": "M", "accent": "Canadian"},
	    282: { "age": 23, "gender": "F", "accent": "English"},
	    362: { "age": 29, "gender": "F", "accent": "American"},
	    294: { "age": 33, "gender": "F", "accent": "American"},
	    274: { "age": 22, "gender": "M", "accent": "English"},
	    279: { "age": 23, "gender": "M", "accent": "English"},
	    281: { "age": 29, "gender": "M", "accent": "Scottish"},
	    286: { "age": 23, "gender": "M", "accent": "English"},
	    258: { "age": 22, "gender": "M", "accent": "English"},
	    247: { "age": 22, "gender": "M", "accent": "Scottish"},
	    351: { "age": 21, "gender": "F", "accent": "NorthernIrish"},
	    283: { "age": 24, "gender": "F", "accent": "Irish"},
	    334: { "age": 18, "gender": "M", "accent": "American"},
	    333: { "age": 19, "gender": "F", "accent": "American"},
	    295: { "age": 23, "gender": "F", "accent": "Irish"},
	    330: { "age": 26, "gender": "F", "accent": "American"},
	    335: { "age": 25, "gender": "F", "accent": "NewZealand"},
	    228: { "age": 22, "gender": "F", "accent": "English"},
	    267: { "age": 23, "gender": "F", "accent": "English"},
	    273: { "age": 18, "gender": "F", "accent": "English"}
	}



# Static list of speakers for dropdown
SPEAKER_CHOICES = [
    f"{sid} - {data['gender']} - {data['accent']} (Age {data['age']})"
    for sid, data in SPEAKER_METADATA.items()
]

# VCTK model (multi-speaker)
MODEL_NAME = "tts_models/en/vctk/vits"
tts = TTS(model_name=MODEL_NAME, progress_bar=False, gpu=False)

# Extract plain text from docx, ignoring hyperlinks
def extract_text_ignoring_hyperlinks(docx_file):
    doc = Document(docx_file.name)
    text_blocks = []
    for para in doc.paragraphs:
        # Remove hyperlinks using regex or by inspecting runs
        if para.text.strip():
            clean_text = re.sub(r'https?://\S+', '', para.text)
            text_blocks.append(clean_text.strip())
    return text_blocks

# Generate sample audio for preview
def generate_sample_audio(sample_text, selected_speaker):
    if not sample_text.strip():
        raise gr.Error("Sample text cannot be empty.")
    sid = selected_speaker.split(" ")[0]  # Extract speaker ID
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp_wav:
        tts.tts_to_file(text=sample_text, speaker="p"+sid, file_path=tmp_wav.name)
        return tmp_wav.name

# Main conversion function
def docx_to_zipped_wavs(doc_file, selected_speaker):
    sid = selected_speaker.split(" ")[0]
    paragraphs = extract_text_ignoring_hyperlinks(doc_file)
    audio_files = []
    try:
        for i, para in enumerate(paragraphs):
            if not para:
                continue
            with tempfile.NamedTemporaryFile(suffix=f"_{i}.wav", delete=False) as tmp_wav:
                tts.tts_to_file(text=para, speaker="p"+sid, file_path=tmp_wav.name)
                audio_files.append(tmp_wav.name)
    except Exception as e:
        print("Connection interrupted, returning partial result.", str(e))

    # Zip the results
    zip_buffer = BytesIO()
    with zipfile.ZipFile(zip_buffer, "w") as zipf:
        for wav_path in audio_files:
            zipf.write(wav_path, arcname=os.path.basename(wav_path))
    zip_buffer.seek(0)

    # Save the zip temporarily for download
    final_zip = tempfile.NamedTemporaryFile(delete=False, suffix=".zip")
    final_zip.write(zip_buffer.read())
    final_zip.close()
    return final_zip.name

# Gradio UI
with gr.Blocks() as interface:
    gr.Markdown("""# Multi-Paragraph Voiceover Generator
Upload a `.docx` file and convert each paragraph to audio. You can also try a short sample first.
""")

    with gr.Row():
        sample_text = gr.Textbox(label="Sample Text (Max 500 chars)", max_lines=4, lines=3, max_length=500)
        speaker_dropdown = gr.Dropdown(label="Select Speaker", choices=SPEAKER_CHOICES, value=SPEAKER_CHOICES[0])

    # sample_button = gr.Button("Generate Sample Audio")
    # sample_audio = gr.Audio(label="Sample Audio", type="filepath")

    with gr.Row():
        docx_input = gr.File(label="Upload .docx File", type="filepath")
        convert_button = gr.Button("Generate WAV Zip")

    final_output = gr.File(label="Download ZIP of WAVs")

    # sample_button.click(fn=generate_sample_audio, inputs=[sample_text, speaker_dropdown], outputs=sample_audio)
    convert_button.click(fn=docx_to_zipped_wavs, inputs=[docx_input, speaker_dropdown], outputs=final_output)

if __name__ == "__main__":
    interface.launch()