app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from snac import SNAC
import soundfile as sf
import numpy as np
import tempfile
import spaces

# Load models at startup
print("Loading Maya1 model...")
model = AutoModelForCausalLM.from_pretrained(
    "maya-research/maya1", 
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(
    "maya-research/maya1",
    trust_remote_code=True
)
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

print("Loading SNAC audio decoder...")
snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz").eval()

@spaces.GPU(duration=180)
def generate_speech(text, voice_description):
    """Generate speech from text using Maya1 model with ZeroGPU"""
    
    if not text.strip():
        raise gr.Error("Please enter some text to convert to speech!")
    
    if not voice_description.strip():
        voice_description = "Realistic voice with neutral tone and conversational pacing."
    
    try:
        # Move models to GPU
        device = "cuda" if torch.cuda.is_available() else "cpu"
        model.to(device)
        snac_model.to(device)
        
        # Create prompt
        prompt = f'<description="{voice_description}"> {text}'
        
        # Tokenize input
        inputs = tokenizer(prompt, return_tensors="pt").to(device)
        
        # Generate
        with torch.inference_mode():
            outputs = model.generate(
                input_ids=inputs['input_ids'],
                attention_mask=inputs.get('attention_mask', None),
                max_new_tokens=1000, 
                temperature=0.7, 
                top_p=0.9, 
                do_sample=True,
                pad_token_id=tokenizer.pad_token_id,
                eos_token_id=None,
            )
        
        # Extract SNAC audio tokens
        generated_ids = outputs[0, inputs['input_ids'].shape[1]:]
        snac_tokens = [t.item() for t in generated_ids if 128266 <= t <= 156937]
        
        if len(snac_tokens) < 7:
            raise gr.Error(
                f"Not enough audio tokens generated ({len(snac_tokens)}). "
                f"Try using longer text or different voice description."
            )
        
        # Decode SNAC tokens to audio frames
        frames = len(snac_tokens) // 7
        codes = [[], [], []]
        
        for i in range(frames):
            s = snac_tokens[i*7:(i+1)*7]
            codes[0].append((s[0]-128266) % 4096)
            codes[1].extend([(s[1]-128266) % 4096, (s[4]-128266) % 4096])
            codes[2].extend([
                (s[2]-128266) % 4096, 
                (s[3]-128266) % 4096, 
                (s[5]-128266) % 4096, 
                (s[6]-128266) % 4096
            ])
        
        # Generate final audio with SNAC decoder
        codes_tensor = [
            torch.tensor(c, dtype=torch.long, device=device).unsqueeze(0) 
            for c in codes
        ]
        
        with torch.inference_mode():
            audio = snac_model.decoder(
                snac_model.quantizer.from_codes(codes_tensor)
            )[0, 0].cpu().numpy()
        
        # Save to temporary file
        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as f:
            sf.write(f.name, audio, 24000)
            return f.name
            
    except Exception as e:
        import traceback
        traceback.print_exc()
        raise gr.Error(f"Error generating speech: {str(e)}")

# Examples
examples = [
    [
        "Hello! This is Maya1 <laugh> the best open source voice AI model with emotions.",
        "Realistic male voice in the 30s age with american accent. Normal pitch, warm timbre, conversational pacing."
    ],
    [
        "I'm so excited to share this amazing news with you! This is incredible and wonderful!",
        "Energetic female voice with enthusiastic tone. Higher pitch, bright timbre, upbeat pacing."
    ],
    [
        "In a world of constant change, one thing remains certain: the power of human connection and understanding.",
        "Deep male voice with authoritative tone. Low pitch, resonant timbre, steady pacing."
    ],
]

# Create Gradio interface
with gr.Blocks(theme=gr.themes.Soft(), title="Maya1 Text-to-Speech") as demo:
    gr.Markdown(
        """
        # 🎙️ Maya1 Text-to-Speech
        Generate emotional and realistic speech with natural language voice design
        
        Built with [anycoder](https://huggingface.co/spaces/akhaliq/anycoder)
        """
    )
    
    with gr.Row():
        with gr.Column():
            text_input = gr.Textbox(
                label="Text to Speak",
                placeholder="Enter your text here... You can use <laugh>, <sigh>, and other emotion tags!",
                lines=5,
                value="Hello! This is Maya1 <laugh> the best open source voice AI model with emotions."
            )
            
            voice_description = gr.Textbox(
                label="Voice Description",
                placeholder="Describe the voice characteristics (age, gender, accent, pitch, timbre, pacing)...",
                lines=3,
                value="Realistic male voice in the 30s age with american accent. Normal pitch, warm timbre, conversational pacing."
            )
            
            generate_btn = gr.Button("🎤 Generate Speech", variant="primary", size="lg")
        
        with gr.Column():
            audio_output = gr.Audio(
                label="Generated Speech",
                type="filepath"
            )
            
            gr.Markdown("""
            ### 💡 Tips
            - Use emotion tags: `<laugh>`, `<sigh>`, `<whisper>`, `<shout>`
            - Describe voice with: age, gender, accent, pitch, timbre, pacing
            - Longer text works better (20+ words recommended)
            
            ### About
            Maya1 is an open-source voice AI model that generates realistic, emotional speech from text using natural language voice descriptions.
            """)
    
    # Generate speech button
    generate_btn.click(
        fn=generate_speech,
        inputs=[text_input, voice_description],
        outputs=[audio_output]
    )
    
    # Examples section
    gr.Examples(
        examples=examples,
        inputs=[text_input, voice_description],
        outputs=[audio_output],
        fn=generate_speech,
        cache_examples=False,
        label="Example Prompts"
    )

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