gemini_tts_api.py

#!/usr/bin/env python3
"""
Gemini TTS API Server
A FastAPI-based REST API for Google's Gemini Text-to-Speech service
with concurrent request handling and audio format conversion.
"""

import os
import asyncio
import json
import base64
import uuid
from datetime import datetime
from typing import Optional, List, Dict, Any
from io import BytesIO
import aiohttp
import aiofiles
from fastapi import FastAPI, HTTPException, BackgroundTasks, Request
from fastapi.responses import FileResponse, JSONResponse
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field
from pydub import AudioSegment
import uvicorn

# Pydantic models for request/response
class VoiceConfig(BaseModel):
    voice_name: str = Field(default="Zephyr", description="Voice name (e.g., Zephyr, Puck)")

class SpeakerConfig(BaseModel):
    speaker: str = Field(description="Speaker identifier")
    voice_config: VoiceConfig

class TTSRequest(BaseModel):
    text: str = Field(description="Text to convert to speech")
    speakers: Optional[List[SpeakerConfig]] = Field(
        default=None, 
        description="Multi-speaker configuration (optional)"
    )
    voice_name: Optional[str] = Field(
        default="Zephyr", 
        description="Single voice name (used if speakers not provided)"
    )
    output_format: str = Field(default="wav", description="Output format: wav or mp3")
    speed_factor: float = Field(default=1.0, description="Speed adjustment factor")
    temperature: float = Field(default=1.0, description="Generation temperature")

class TTSResponse(BaseModel):
    task_id: str
    status: str
    message: str
    audio_url: Optional[str] = None
    metadata: Optional[Dict[str, Any]] = None

class TaskStatus(BaseModel):
    task_id: str
    status: str
    progress: Optional[str] = None
    error: Optional[str] = None
    result: Optional[Dict[str, Any]] = None

# Global task storage (in production, use Redis or database)
tasks: Dict[str, Dict[str, Any]] = {}

# FastAPI app initialization
app = FastAPI(
    title="Gemini TTS API",
    description="Text-to-Speech API using Google's Gemini model with concurrent request handling",
    version="1.0.0"
)

# CORS middleware
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],  # Configure appropriately for production
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Configuration
def get_api_keys():
    """Get API keys from environment variables"""
    # Support multiple formats for API keys
    api_keys = []
    
    # Single API key (backward compatibility)
    single_key = os.getenv('GEMINI_API_KEY')
    if single_key:
        api_keys.append(single_key.strip())
    
    # Multiple API keys (comma-separated)
    multi_keys = os.getenv('GEMINI_API_KEYS')
    if multi_keys:
        keys = [key.strip() for key in multi_keys.split(',') if key.strip()]
        api_keys.extend(keys)
    
    # Individual API keys (GEMINI_API_KEY_1, GEMINI_API_KEY_2, etc.)
    i = 1
    while True:
        key = os.getenv(f'GEMINI_API_KEY_{i}')
        if not key:
            break
        api_keys.append(key.strip())
        i += 1
    
    # Remove duplicates while preserving order
    seen = set()
    unique_keys = []
    for key in api_keys:
        if key not in seen:
            seen.add(key)
            unique_keys.append(key)
    
    return unique_keys

GEMINI_API_KEYS = get_api_keys()
MODEL_ID = "gemini-2.5-flash-preview-tts"
GENERATE_CONTENT_API = "streamGenerateContent"
OUTPUT_DIR = "/tmp/audio_files"
MAX_CONCURRENT_REQUESTS = 10
RATE_LIMIT_RETRY_DELAY = 60  # seconds to wait after rate limit
MAX_RETRIES_PER_KEY = 2

# API key management
class APIKeyManager:
    def __init__(self, api_keys: List[str]):
        self.api_keys = api_keys
        self.current_key_index = 0
        self.key_stats = {key: {"requests": 0, "failures": 0, "last_rate_limit": None} for key in api_keys}
        self.lock = asyncio.Lock()
    
    async def get_next_key(self) -> Optional[str]:
        """Get the next available API key"""
        async with self.lock:
            if not self.api_keys:
                return None
            
            # Try to find a key that's not rate limited
            for _ in range(len(self.api_keys)):
                key = self.api_keys[self.current_key_index]
                stats = self.key_stats[key]
                
                # Check if this key is currently rate limited
                if stats["last_rate_limit"]:
                    time_since_limit = datetime.now().timestamp() - stats["last_rate_limit"]
                    if time_since_limit < RATE_LIMIT_RETRY_DELAY:
                        # Still rate limited, try next key
                        self.current_key_index = (self.current_key_index + 1) % len(self.api_keys)
                        continue
                    else:
                        # Rate limit period has passed, reset
                        stats["last_rate_limit"] = None
                
                # This key is available
                stats["requests"] += 1
                return key
            
            # All keys are rate limited, return the one with oldest rate limit
            oldest_key = min(
                self.api_keys, 
                key=lambda k: self.key_stats[k]["last_rate_limit"] or 0
            )
            return oldest_key
    
    async def mark_rate_limited(self, api_key: str):
        """Mark an API key as rate limited"""
        async with self.lock:
            if api_key in self.key_stats:
                self.key_stats[api_key]["last_rate_limit"] = datetime.now().timestamp()
                self.key_stats[api_key]["failures"] += 1
    
    async def mark_success(self, api_key: str):
        """Mark an API key as successful (reset failure count)"""
        async with self.lock:
            if api_key in self.key_stats:
                self.key_stats[api_key]["failures"] = max(0, self.key_stats[api_key]["failures"] - 1)
    
    def get_stats(self) -> dict:
        """Get statistics for all API keys"""
        return {
            "total_keys": len(self.api_keys),
            "key_stats": self.key_stats.copy()
        }

# Initialize API key manager
api_key_manager = APIKeyManager(GEMINI_API_KEYS)

# Ensure output directory exists
os.makedirs(OUTPUT_DIR, exist_ok=True)

# Semaphore to limit concurrent requests
semaphore = asyncio.Semaphore(MAX_CONCURRENT_REQUESTS)

async def convert_and_adjust_audio(
    audio_data: bytes, 
    output_format: str = "wav", 
    speed_factor: float = 1.0
) -> tuple[bytes, str]:
    """
    Convert PCM audio data to specified format and adjust speed asynchronously
    """
    def _convert():
        # Create AudioSegment from raw PCM data
        audio = AudioSegment(
            data=audio_data,
            sample_width=2,  # 16-bit = 2 bytes
            frame_rate=24000,  # 24kHz
            channels=1  # mono
        )
        
        # Adjust speed by changing frame rate
        if speed_factor != 1.0:
            new_frame_rate = int(audio.frame_rate * speed_factor)
            audio_speed_adjusted = audio._spawn(
                audio.raw_data, 
                overrides={"frame_rate": new_frame_rate}
            )
            audio_speed_adjusted = audio_speed_adjusted.set_frame_rate(audio.frame_rate)
        else:
            audio_speed_adjusted = audio
        
        # Export to desired format
        buffer = BytesIO()
        if output_format.lower() == "mp3":
            audio_speed_adjusted.export(buffer, format="mp3", bitrate="128k")
            return buffer.getvalue(), "mp3"
        else:
            audio_speed_adjusted.export(buffer, format="wav")
            return buffer.getvalue(), "wav"
    
    # Run audio processing in thread pool to avoid blocking
    loop = asyncio.get_event_loop()
    return await loop.run_in_executor(None, _convert)

async def generate_tts_audio(
    task_id: str,
    text: str,
    speakers: Optional[List[SpeakerConfig]] = None,
    voice_name: str = "Zephyr",
    output_format: str = "wav",
    speed_factor: float = 1.0,
    temperature: float = 1.0
):
    """
    Generate TTS audio using Gemini API with multiple API key support and rate limit handling
    """
    async with semaphore:  # Limit concurrent requests
        try:
            # Update task status
            tasks[task_id]["status"] = "processing"
            tasks[task_id]["progress"] = "Preparing request"
            
            # Prepare request data
            if speakers:
                # Multi-speaker configuration
                speech_config = {
                    "multi_speaker_voice_config": {
                        "speaker_voice_configs": [
                            {
                                "speaker": speaker.speaker,
                                "voice_config": {
                                    "prebuilt_voice_config": {
                                        "voice_name": speaker.voice_config.voice_name
                                    }
                                }
                            }
                            for speaker in speakers
                        ]
                    }
                }
            else:
                # Single voice configuration
                speech_config = {
                    "voice_config": {
                        "prebuilt_voice_config": {
                            "voice_name": voice_name
                        }
                    }
                }
            
            request_data = {
                "contents": [
                    {
                        "role": "user",
                        "parts": [{"text": text}]
                    }
                ],
                "generationConfig": {
                    "responseModalities": ["audio"],
                    "temperature": temperature,
                    "speech_config": speech_config
                }
            }
            
            # API endpoint
            url = f"https://generativelanguage.googleapis.com/v1beta/models/{MODEL_ID}:{GENERATE_CONTENT_API}"
            
            tasks[task_id]["progress"] = "Calling Gemini API"
            
            # Try multiple API keys with rate limit handling
            last_error = None
            attempts = 0
            max_total_attempts = len(GEMINI_API_KEYS) * MAX_RETRIES_PER_KEY if GEMINI_API_KEYS else 1
            
            while attempts < max_total_attempts:
                current_api_key = await api_key_manager.get_next_key()
                if not current_api_key:
                    raise HTTPException(status_code=500, detail="No API keys available")
                
                attempts += 1
                tasks[task_id]["progress"] = f"Attempting API call (attempt {attempts}/{max_total_attempts})"
                
                try:
                    # Make async API request
                    async with aiohttp.ClientSession() as session:
                        async with session.post(
                            url,
                            headers={"Content-Type": "application/json"},
                            params={"key": current_api_key},
                            json=request_data,
                            timeout=aiohttp.ClientTimeout(total=120)  # 2 minute timeout
                        ) as response:
                            
                            # Handle different HTTP status codes
                            if response.status == 200:
                                # Success! Mark key as successful and proceed
                                await api_key_manager.mark_success(current_api_key)
                                response_data = await response.json()
                                break
                                
                            elif response.status == 429:  # Rate limit exceeded
                                error_text = await response.text()
                                await api_key_manager.mark_rate_limited(current_api_key)
                                last_error = f"Rate limit exceeded for API key: {error_text}"
                                print(f"Rate limit hit for key ending in ...{current_api_key[-4:]}, trying next key")
                                continue
                                
                            elif response.status in [403, 401]:  # Auth errors
                                error_text = await response.text()
                                await api_key_manager.mark_rate_limited(current_api_key)  # Temporarily disable this key
                                last_error = f"Authentication error: {error_text}"
                                print(f"Auth error for key ending in ...{current_api_key[-4:]}: {error_text}")
                                continue
                                
                            else:  # Other HTTP errors
                                error_text = await response.text()
                                last_error = f"HTTP {response.status}: {error_text}"
                                # Don't mark as rate limited for other errors, but still try next key
                                continue
                
                except asyncio.TimeoutError:
                    last_error = "Request timeout"
                    continue
                except aiohttp.ClientError as e:
                    last_error = f"Client error: {str(e)}"
                    continue
                except Exception as e:
                    last_error = f"Unexpected error: {str(e)}"
                    continue
            
            else:
                # All attempts failed
                raise HTTPException(
                    status_code=500, 
                    detail=f"All API keys exhausted. Last error: {last_error}"
                )
            
            tasks[task_id]["progress"] = "Processing audio data"
            
            # Extract audio data
            if response_data and len(response_data) > 0:
                candidates = response_data[0].get("candidates", [])
                if not candidates:
                    raise HTTPException(status_code=500, detail="No candidates in response")
                
                parts = candidates[0].get("content", {}).get("parts", [])
                audio_data_b64 = None
                
                for part in parts:
                    if "inlineData" in part:
                        audio_data_b64 = part["inlineData"].get("data", "")
                        break
                
                if not audio_data_b64:
                    raise HTTPException(status_code=500, detail="No audio data found in response")
                
                # Decode base64 audio data
                audio_data = base64.b64decode(audio_data_b64)
                
                tasks[task_id]["progress"] = "Converting audio format"
                
                # Convert and adjust audio
                converted_audio, file_ext = await convert_and_adjust_audio(
                    audio_data, output_format, speed_factor
                )
                
                # Generate filename
                timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
                filename = f"gemini_audio_{task_id}_{timestamp}.{file_ext}"
                filepath = os.path.join(OUTPUT_DIR, filename)
                
                # Save audio file
                async with aiofiles.open(filepath, "wb") as f:
                    await f.write(converted_audio)
                
                # Update task with results
                tasks[task_id].update({
                    "status": "completed",
                    "progress": "Completed",
                    "result": {
                        "filename": filename,
                        "filepath": filepath,
                        "format": output_format.upper(),
                        "speed_factor": speed_factor,
                        "original_size": len(audio_data),
                        "converted_size": len(converted_audio),
                        "audio_url": f"/audio/{filename}"
                    }
                })
                
        except Exception as e:
            tasks[task_id].update({
                "status": "failed",
                "error": str(e)
            })

# API Endpoints

@app.get("/")
async def root():
    """Root endpoint with API information"""
    return {
        "message": "Gemini TTS API Server",
        "version": "1.0.0",
        "endpoints": {
            "POST /tts": "Generate TTS audio",
            "GET /status/{task_id}": "Get task status",
            "GET /audio/{filename}": "Download audio file",
            "GET /tasks": "List all tasks"
        }
    }

@app.post("/tts", response_model=TTSResponse)
async def create_tts_task(
    request: TTSRequest,
    background_tasks: BackgroundTasks
):
    """
    Create a new TTS generation task
    """
    if not GEMINI_API_KEYS:
        raise HTTPException(status_code=500, detail="No GEMINI_API_KEYs configured. Please set GEMINI_API_KEY, GEMINI_API_KEYS, or GEMINI_API_KEY_1, GEMINI_API_KEY_2, etc.")
    
    # Generate unique task ID
    task_id = str(uuid.uuid4())
    
    # Initialize task
    tasks[task_id] = {
        "task_id": task_id,
        "status": "queued",
        "created_at": datetime.now().isoformat(),
        "request": request.dict()
    }
    
    # Start background task
    background_tasks.add_task(
        generate_tts_audio,
        task_id,
        request.text,
        request.speakers,
        request.voice_name,
        request.output_format,
        request.speed_factor,
        request.temperature
    )
    
    return TTSResponse(
        task_id=task_id,
        status="queued",
        message="TTS generation task created successfully"
    )

@app.get("/status/{task_id}", response_model=TaskStatus)
async def get_task_status(task_id: str):
    """
    Get the status of a TTS generation task
    """
    if task_id not in tasks:
        raise HTTPException(status_code=404, detail="Task not found")
    
    task = tasks[task_id]
    return TaskStatus(
        task_id=task_id,
        status=task["status"],
        progress=task.get("progress"),
        error=task.get("error"),
        result=task.get("result")
    )

@app.get("/audio/{filename}")
async def download_audio(filename: str):
    """
    Download generated audio file
    """
    filepath = os.path.join(OUTPUT_DIR, filename)
    if not os.path.exists(filepath):
        raise HTTPException(status_code=404, detail="Audio file not found")
    
    return FileResponse(
        filepath,
        media_type="application/octet-stream",
        filename=filename
    )

@app.get("/tasks")
async def list_tasks():
    """
    List all tasks with their current status
    """
    return {"tasks": list(tasks.values())}

@app.delete("/tasks/{task_id}")
async def delete_task(task_id: str):
    """
    Delete a task and its associated audio file
    """
    if task_id not in tasks:
        raise HTTPException(status_code=404, detail="Task not found")
    
    task = tasks[task_id]
    
    # Delete audio file if it exists
    if task.get("result") and task["result"].get("filepath"):
        filepath = task["result"]["filepath"]
        if os.path.exists(filepath):
            os.remove(filepath)
    
    # Remove task from memory
    del tasks[task_id]
    
    return {"message": "Task deleted successfully"}

@app.get("/health")
async def health_check():
    """
    Health check endpoint with API key status
    """
    api_stats = api_key_manager.get_stats()
    
    return {
        "status": "healthy",
        "timestamp": datetime.now().isoformat(),
        "active_tasks": len([t for t in tasks.values() if t["status"] in ["queued", "processing"]]),
        "total_tasks": len(tasks),
        "api_keys": {
            "total_configured": api_stats["total_keys"],
            "available_keys": len([
                key for key, stats in api_stats["key_stats"].items()
                if not stats["last_rate_limit"] or 
                (datetime.now().timestamp() - stats["last_rate_limit"]) > RATE_LIMIT_RETRY_DELAY
            ]),
            "rate_limited_keys": len([
                key for key, stats in api_stats["key_stats"].items()
                if stats["last_rate_limit"] and 
                (datetime.now().timestamp() - stats["last_rate_limit"]) <= RATE_LIMIT_RETRY_DELAY
            ])
        }
    }

@app.get("/api-keys/stats")
async def get_api_key_stats():
    """
    Get detailed statistics for all API keys
    """
    stats = api_key_manager.get_stats()
    
    # Mask API keys for security (show only last 4 characters)
    masked_stats = {}
    for key, data in stats["key_stats"].items():
        masked_key = f"***{key[-4:]}" if len(key) > 4 else "***"
        masked_stats[masked_key] = {
            **data,
            "is_rate_limited": (
                data["last_rate_limit"] and 
                (datetime.now().timestamp() - data["last_rate_limit"]) <= RATE_LIMIT_RETRY_DELAY
            ) if data["last_rate_limit"] else False,
            "time_until_available": max(0, RATE_LIMIT_RETRY_DELAY - (
                datetime.now().timestamp() - data["last_rate_limit"]
            )) if data["last_rate_limit"] else 0
        }
    
    return {
        "total_keys": stats["total_keys"],
        "key_statistics": masked_stats,
        "rate_limit_settings": {
            "retry_delay_seconds": RATE_LIMIT_RETRY_DELAY,
            "max_retries_per_key": MAX_RETRIES_PER_KEY
        }
    }

# Cleanup old files periodically (you might want to implement this with a proper scheduler)
async def cleanup_old_files():
    """
    Clean up old audio files and completed tasks
    """
    # This is a simple implementation - consider using APScheduler for production
    pass

if __name__ == "__main__":
    # Configuration for running the server
    uvicorn.run(
        "gemini_tts_api:app",
        host="0.0.0.0",
        port=8000,
        reload=True,  # Set to False in production
        workers=1  # Use multiple workers in production with proper task storage
    )