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):
    voice1 ="vi-VN-NamMinhNeural - vi-VN (Male)"  #good for reading
    voice1F ="en-US-EmmaMultilingualNeural - en-US (Female)"
    voice2 ="ko-KR-HyunsuMultilingualNeural - ko-KR (Male)"
    voice2F ="de-DE-SeraphinaMultilingualNeural - de-DE (Female)"
    voice3 = "pt-BR-ThalitaMultilingualNeural - pt-BR (Female)" #Child
    voice_it_M = "it-IT-GiuseppeMultilingualNeural - it-IT (Male)"
    voice_de_M = "de-DE-FlorianMultilingualNeural - de-DE (Male)"
    voice_fr_M = "fr-FR-RemyMultilingualNeural - fr-FR (Male)"
    voice_fr_F = "fr-FR-VivienneMultilingualNeural - fr-FR (Female)"
    voice_en_US_M = "en-US-BrianMultilingualNeural - en-US (Male)"
    voice_en_US_F2 = "en-US-AvaMultilingualNeural - en-US (Female)"

    if not text.strip():
        return None

    prefix_pattern = re.compile(r"^(1F|1M|2F|2M|1C|3M|3F)([-]?\d*)")
    match = prefix_pattern.match(text)
    voice_short_name = (voice or default_voice).split(" - ")[0]
    extracted_pitch = None
    text2 = text

    if match:
        prefix = match.group(1)
        pitch_mod_str = match.group(2)
        text2 = text[len(match.group(0)):]  # Remove the prefix and optional pitch modifier
        print(f">>>Processing text: '{text}'")
        if prefix == "1F":
            voice_short_name = voice1F.split(" - ")[0]
        elif prefix == "1M":
            voice_short_name = voice1.split(" - ")[0]
        elif prefix == "2F":
            voice_short_name = voice2F.split(" - ")[0]
        elif prefix == "2M":
            voice_short_name = voice2.split(" - ")[0]
        elif prefix == "1C":
            voice_short_name = voice3.split(" - ")[0]
        elif prefix == "3M":
            voice_short_name = voice_it_M.split(" - ")[0] # Using Italian Male for 3M
        elif prefix == "3F":
            voice_short_name = voice_fr_F.split(" - ")[0] # Using French Female for 3F

        if pitch_mod_str:
            try:
                extracted_pitch = int(pitch_mod_str)
            except ValueError:
                print(f"Warning: Invalid pitch modifier '{pitch_mod_str}'")

    rate_str = f"{rate:+d}%"
    current_pitch = pitch
    if extracted_pitch is not None:
        current_pitch += extracted_pitch
        print(f"Applying pitch modification: {extracted_pitch}Hz, new pitch: {current_pitch}Hz")
    elif (voice_short_name == voice3.split(" - ")[0]):
        current_pitch = 70  # Default pitch for the child voice

    pitch_str = f"{current_pitch:+d}Hz"
    try:
        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
    except Exception as e:
        print(f"Edge TTS error processing '{processed_text}': {e}")
        return None        
    #return None 

# 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 quote marks
    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 = "vi-VN-HoaiMyNeural - vi-VN (Female)"  # 👈 Pick one of the available voices
    description = """
    Default = vi-VN-HoaiMyNeural - vi-VN (Female),
    other voices 1F:en-US-EmmaMultilingualNeural - en-US (Female),
                  1M:vi-VN-NamMinhNeural - vi-VN (Male),
                  2F:de-DE-SeraphinaMultilingualNeural - de-DE (Female),
                  2M:ko-KR-HyunsuMultilingualNeural - ko-KR (Male),
                  1C:pt-BR-ThalitaMultilingualNeural - pt-BR (Female),
                  3M:it-IT-GiuseppeMultilingualNeural - it-IT (Male),
                  3F:fr-FR-VivienneMultilingualNeural - fr-FR (Female)
    You can add a pitch modifier after the voice prefix (e.g., 1M-15 for +15Hz pitch).
    Enter your text, select a voice, and adjust the speech rate and pitch.
    The application will process your text segment by segment based on quote marks.
    """
    demo = gr.Interface(
        fn=tts_interface,
        inputs=[
            gr.Textbox(label="Input Text", lines=5, placeholder='Separate dialogue with quote marks. Add voice prefix (e.g., 1M, 1F) before dialogue. You can also add a pitch modifier like 1M-20 "Hello!"'),
            gr.Dropdown(choices=[""] + list(voices.keys()), label="Select Default Voice", value=default_voice),
            gr.Slider(minimum=-50, maximum=50, value=0, label="Default Speech Rate Adjustment (%)", step=1),
            gr.Slider(minimum=-20, maximum=100, value=0, label="Default Pitch Adjustment (Hz)", step=1)
        ],
        outputs=[
            gr.Audio(label="Generated Audio", type="filepath"),
            gr.Markdown(label="Warning", visible=False)
        ],
        title="Vietnamese TTS - all AI voices & pitch changes",
        description=description,
        article="Process text segments with voice prefixes and optional pitch modifiers.",
        analytics_enabled=False,
        allow_flagging=False
    )
    return demo

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