app-divide-text.py

import gradio as gr
import requests
import os
from deployment_options import voice_id_2_name, defualt_values

import uuid

ENDPOINT_URL = "https://sentivue-endpoint.hf.space/v1/tts"
ENDPOINT_TOKEN = os.getenv("endpoint_READ")

print(f"Public demo will call endpoint: {ENDPOINT_URL}")
print(f"Token loaded: {'Yes' if ENDPOINT_TOKEN else 'No'}")

voice_names = list(voice_id_2_name.values())
def generate_speech(text: str, voice_name: str):
    """
    Calls the private FastAPI endpoint and returns audio
    """
    if not text.strip():
        return None, "Please enter some text"
    
    if not ENDPOINT_TOKEN:
        return None, "Error: endpoint_READ token not found in environment"

    try:

        voice_name_2_id = {}
        for vid, name in voice_id_2_name.items():
            voice_name_2_id[name] = vid

        voice_id = voice_name_2_id[voice_name]
        
        payload = {
            "text": text
        }

        print(f"Sending request to: {ENDPOINT_URL}/{voice_id}")
        print(f"Payload: {payload}")

        response = requests.post(
            f"{ENDPOINT_URL}/{voice_id}",
            headers={
                "Authorization": f"Bearer {ENDPOINT_TOKEN}",
                "Content-Type": "application/json"
            },
            json=payload,
            # timeout=60,
            stream=True
        )

        response.raise_for_status()

        # # Return raw WAV bytes - Gradio handles the rest
        # return response.content, "Success!"

        # # Save to temporary WAV file
        # with tempfile.NamedTemporaryFile(delete=False, suffix='.wav') as tmp_file:
        #     tmp_file.write(response.content)
        #     tmp_path = tmp_file.name
        
        # return tmp_path, "Success!"

        # Save to a regular file in current directory (not temp)
        # Generate unique ID for output file
        generation_id = str(uuid.uuid4())[:15]
        output_path = f"speech_{voice_id}_{generation_id}.wav"
        
        with open(output_path, 'wb') as f:
            for chunk in response.iter_content(chunk_size=8192):
                if chunk:
                    f.write(chunk)
        
        return output_path, "Success!"

    except requests.exceptions.RequestException as e:
        error_msg = f"Error calling endpoint: {str(e)}"
        print(error_msg)
        return None, error_msg
    except Exception as e:
        error_msg = f"Unexpected error: {str(e)}"
        print(error_msg)
        return None, error_msg


# ── Gradio Interface ────────────────────────────────────────────────────────

with gr.Blocks(title="pt-PT TTS - Demo") as demo:
    gr.Markdown("""
    # pt-PT Text-to-Speech Demo

    High-quality, natural-sounding speech synthesis for European Portuguese (pt-PT) with human-like prosody and accurate number pronunciation.

    ### Voices
    - **Current:** André  
    - **More voices:** Coming soon, with **extended emotion and prosody control via tags**
    
    """)

    with gr.Row():
        with gr.Column(scale=3):
            text_input = gr.Textbox(
                label="Text to speak",
                placeholder="Write something here...",
                lines=4
            )

            voice_dropdown = gr.Dropdown(
                choices=voice_names,
                value=defualt_values['voice_name'],
                label="Voice"
            )

            submit_btn = gr.Button("Generate Speech", variant="primary")

        with gr.Column(scale=2):
            audio_output = gr.Audio(
                label="Generated Speech",
                autoplay=False
            )

            status_text = gr.Textbox(label="Status", interactive=False)

    submit_btn.click(
        fn=generate_speech,
        inputs=[text_input, voice_dropdown],
        outputs=[audio_output, status_text]
    )

    gr.Markdown("""
                
    ### Info
    - **Language:** pt-PT (European Portuguese only)
    - **Voice:** André
    - **Model size:** ~3B parameters
    - **Architecture:** LLM-based TTS backbone
    - **Training data:** +11k hours of curated pt-PT speech
    - **Inference:** Streaming audio generation via remote endpoint
    
    ### API
    - **Status:** Coming soon  
                                
    ### Fine-tuning
    - **Status:** Coming soon  
    - **Requirements:** ~1.5 hours of recorded speech to fine-tune a custom voice

    The system is designed for natural prosody, accent fidelity, and long-form synthesis in European Portuguese.
    """)

demo.queue().launch()