app.py

import spaces
import gradio as gr
import edge_tts
import asyncio
import tempfile
import os
import re  # Import the regular expression module


# Get all available voices
async def get_voices():
    voices = await edge_tts.list_voices()
    return {f"{v['ShortName']} - {v['Locale']} ({v['Gender']})": v['ShortName'] for v in voices}

# Text-to-speech function for a single paragraph
async def paragraph_to_speech(text, voice, rate, pitch):
    voice3 ="en-US-BrianMultilingualNeural - en-US (Male)"  #good for reading
    voice1F ="en-US-EmmaNeural - en-US (Female)"
    voice2 = "it-IT-GiuseppeMultilingualNeural - it-IT (Male)"
    voice2F = "en-US-JennyNeural - en-US (Female)"      
    voice1 = "en-AU-WilliamNeural - en-AU (Male)" 
    voice3F = "en-HK-YanNeural - en-HK (Female)"
    voice4 = "en-GB-MaisieNeural - en-GB (Female)"  #Child
    if not text.strip():
        return None
    if text.startswith("1F"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice1F.split(" - ")[0]
    elif text.startswith("2F"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice2F.split(" - ")[0] 
    elif text.startswith("3F"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice3F.split(" - ")[0]     
    elif text.startswith("1M"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice1.split(" - ")[0]        
    elif text.startswith("2M"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice2.split(" - ")[0]          
    elif text.startswith("3M"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice3.split(" - ")[0]   
    elif text.startswith("1C"):
        text2 = text[2:]  # Remove the first two characters ("FF")
        voice_short_name =voice4.split(" - ")[0]          
    else:
        # Use selected voice, or fallback to default
        voice_short_name = (voice or default_voice).split(" - ")[0]
        text2=text
    rate_str = f"{rate:+d}%"
    pitch_str = f"{pitch:+d}Hz"
    communicate = edge_tts.Communicate(text2, voice_short_name, rate=rate_str, pitch=pitch_str)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    return tmp_path

# Main text-to-speech function that processes paragraphs
async def text_to_speech(text, voice, rate, pitch):
    if not text.strip():
        return None, gr.Warning("Please enter text to convert.")
    if not voice:
        return None, gr.Warning("Please select a voice.")

    # Split by two or more newline characters, optionally preceded by carriage returns
    #paragraphs = [p for p in re.split(r'\r?\n\r?\n+', text) if p.strip()]
    paragraphs = [p.strip() for p in re.split(r'"', text) if p.strip()]

    audio_files = []
    for paragraph in paragraphs:
        audio_path = await paragraph_to_speech(paragraph, voice, rate, pitch)
        if audio_path:
            audio_files.append(audio_path)

    if not audio_files:
        return None, None  # No audio generated

    # Combine audio files if there are multiple paragraphs
    if len(audio_files) == 1:
        return audio_files[0], None
    else:
        # Simple concatenation for now - consider using a proper audio editing library for smoother transitions
        combined_audio_path = tempfile.mktemp(suffix=".mp3")
        with open(combined_audio_path, 'wb') as outfile:
            for filename in audio_files:
                with open(filename, 'rb') as infile:
                    outfile.write(infile.read())
                os.remove(filename)  # Clean up individual files
        return combined_audio_path, None

# Gradio interface function
@spaces.GPU
def tts_interface(text, voice, rate, pitch):
    audio, warning = asyncio.run(text_to_speech(text, voice, rate, pitch))
    return audio, warning

# Create Gradio application
import gradio as gr

async def create_demo():
    voices = await get_voices()
    default_voice = "en-US-AndrewNeural - en-US (Male)"  # 👈 Pick one of the available voices
    description = """
    Default = male, other voices 1F:US_Emma, 2F:US_Jenny, 3F:HK_Yan, 1M:AU_Will, 2M:IT_Guiseppe,3M:US_Brian,  1C: Childvoice
    Enter your text, select a voice, and adjust the speech rate and pitch.
    The application will process your text paragraph by paragraph (separated by two blank lines).
    """

    demo = gr.Interface(
        fn=tts_interface,
        inputs=[
            gr.Textbox(label="Input Text", lines=5, placeholder="Separate paragraphs with two blank lines."),
            gr.Dropdown(choices=[""] + list(voices.keys()), label="Select Voice", value=default_voice),
            gr.Slider(minimum=-50, maximum=50, value=0, label="Speech Rate Adjustment (%)", step=1),
            gr.Slider(minimum=-20, maximum=20, value=0, label="Pitch Adjustment (Hz)", step=1)
        ],
        outputs=[
            gr.Audio(label="Generated Audio", type="filepath"),
            gr.Markdown(label="Warning", visible=False)
        ],
        title="Voicecloning.be Text-to-Speech (Paragraph by Paragraph)",
        description=description,
        article="Process text paragraph by paragraph for smoother output.",
        analytics_enabled=False,
        allow_flagging=False
    )
    return demo

# Run the application
if __name__ == "__main__":
    demo = asyncio.run(create_demo())
    demo.launch()