import os import io import asyncio import time import shutil import numpy as np import psutil import soundfile as sf import subprocess import tempfile from concurrent.futures import ThreadPoolExecutor from typing import Optional, Generator from contextlib import asynccontextmanager import logging import aiofiles import torch from fastapi import FastAPI, HTTPException, UploadFile, File, Form, Query from fastapi.responses import Response, StreamingResponse from fastapi.middleware.cors import CORSMiddleware from pydantic import BaseModel, Field import re import hashlib from functools import lru_cache # Ensure the cloned neutts-air repository is in the path import sys sys.path.append(os.path.join(os.getcwd(), 'neutts-air')) from neuttsair.neutts import NeuTTSAir # Configure logging logging.basicConfig(level=logging.INFO) logger = logging.getLogger("NeuTTS-API") # --- Configuration & Utility Functions --- # Explicitly use CPU as per Dockerfile and Hugging Face free tier compatibility DEVICE = "cpu" # Configure Max Workers for concurrent synthesis threads (1-2 is safe for CPU-only) MAX_WORKERS = 2 tts_executor = ThreadPoolExecutor(max_workers=MAX_WORKERS) SAMPLE_RATE = 24000 CLEANUP_THRESHOLD = 300 # 1 hour in seconds TEMP_AUDIO_DIR = "temp_audio" GENERATED_AUDIO_DIR = "generated_audio" os.makedirs(TEMP_AUDIO_DIR, exist_ok=True) os.makedirs(GENERATED_AUDIO_DIR, exist_ok=True) class TTSRequestModel(BaseModel): """Model for non-file inputs to synthesis and streaming.""" text: str = Field(..., min_length=1, max_length=1000) speed: float = Field(default=1.0, ge=0.5, le=2.0) output_format: str = Field(default="wav", pattern="^(wav|mp3|flac)$") async def convert_to_wav_in_memory(upload_file: UploadFile) -> io.BytesIO: """ Converts uploaded audio to a 24kHz WAV in memory using FFmpeg pipes. This avoids all intermediate disk I/O for maximum speed. """ ffmpeg_command = [ "ffmpeg", "-i", "pipe:0", # Read from stdin "-f", "wav", "-ar", str(SAMPLE_RATE), "-ac", "1", "-c:a", "pcm_s16le", "pipe:1" # Write to stdout ] # Start the subprocess with pipes for stdin, stdout, and stderr proc = await asyncio.create_subprocess_exec( *ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE ) # Stream the uploaded file data into ffmpeg's stdin # and capture the resulting WAV data from its stdout wav_data, stderr_data = await proc.communicate(input=await upload_file.read()) if proc.returncode != 0: error_message = stderr_data.decode() logger.error(f"In-memory conversion failed: {error_message}") # Provide the last line of the FFmpeg error to the user error_detail = error_message.splitlines()[-1] if error_message else "Unknown FFmpeg error." raise HTTPException(status_code=400, detail=f"Audio format conversion failed: {error_detail}") logger.info("In-memory FFmpeg conversion successful.") # Return the raw WAV data in a BytesIO buffer, ready for the model return io.BytesIO(wav_data) # --- Model Wrapper and Logic --- class NeuTTSWrapper: def __init__(self, device: str = "cpu"): self.tts_model = None self.device = device self.load_model() def load_model(self): try: logger.info(f"Loading NeuTTSAir model on device: {self.device}") # Ensure we respect the CPU configuration self.tts_model = NeuTTSAir(backbone_device=self.device, codec_device=self.device) logger.info("✅ NeuTTSAir model loaded successfully.") except Exception as e: logger.error(f"❌ Model loading failed: {e}") raise def _convert_to_streamable_format(self, audio_data: np.ndarray, audio_format: str) -> bytes: """Converts NumPy audio array to streamable bytes in the specified format.""" audio_buffer = io.BytesIO() try: sf.write(audio_buffer, audio_data, SAMPLE_RATE, format=audio_format) except Exception as e: logger.error(f"Failed to write audio data to format {audio_format}: {e}") raise audio_buffer.seek(0) return audio_buffer.read() def _split_text_into_chunks(self, text: str) -> list[str]: """ Splits text into sentences OR clauses using a robust regex. This is fast, library-free, and now handles commas. """ # This regex now finds all sequences of characters that are not a sentence-ending # or clause-ending punctuation mark, followed by that punctuation. # The only change is adding ',' to the character sets. chunks = re.findall(r'[^.,!?]+[.,!?]*', text) return [c.strip() for c in chunks if c.strip()] @lru_cache(maxsize=32) def _get_or_create_reference_encoding(self, audio_content_hash: str, audio_bytes: bytes) -> torch.Tensor: """ Caches the expensive reference encoding operation using an in-memory LRU cache. The hash of the audio content is the key. """ logger.info(f"Cache miss for hash: {audio_content_hash[:10]}... Encoding new reference.") # The model's encode_reference can take a file-like object (BytesIO) return self.tts_model.encode_reference(io.BytesIO(audio_bytes)) def generate_speech_blocking(self, text: str, ref_audio_bytes: bytes, reference_text: str) -> np.ndarray: """Blocking synthesis using cached reference encoding.""" # 1. Hash the audio bytes to get a cache key audio_hash = hashlib.sha256(ref_audio_bytes).hexdigest() # 2. Get the encoding from the cache (or create it if new) ref_s = self._get_or_create_reference_encoding(audio_hash, ref_audio_bytes) # 3. Infer full text with torch.no_grad(): audio = self.tts_model.infer(text, ref_s, reference_text) return audio def stream_speech_blocking(self, text: str, ref_audio_bytes: bytes, reference_text: str, speed: float, audio_format: str) -> Generator[bytes, None, None]: """Sentence-by-Sentence Streaming using cached reference encoding.""" logger.info(f"Starting streaming synthesis for text length: {len(text)}") # 1. Hash the audio bytes once audio_hash = hashlib.sha256(ref_audio_bytes).hexdigest() # 2. Get the reference encoding from cache, once for the whole stream ref_s = self._get_or_create_reference_encoding(audio_hash, ref_audio_bytes) # 3. Split text using the new regex method sentences = self._split_text_into_chunks(text) # 4. Stream chunks for i, sentence in enumerate(sentences): if not sentence.strip(): continue logger.debug(f"Generating streaming chunk {i+1}: '{sentence[:30]}...'") with torch.no_grad(): audio_chunk = self.tts_model.infer(sentence, ref_s, reference_text) yield self._convert_to_streamable_format(audio_chunk, audio_format) logger.info("Streaming synthesis complete.") # --- Asynchronous Offloading --- async def run_blocking_task_async(func, *args, **kwargs): """Offloads a blocking function call to the ThreadPoolExecutor.""" loop = asyncio.get_event_loop() return await loop.run_in_executor( tts_executor, lambda: func(*args, **kwargs) ) async def save_upload_file_async(upload_file: UploadFile) -> str: """Asynchronously saves the UploadFile to disk.""" temp_filename = os.path.join(TEMP_AUDIO_DIR, f"{time.time()}_{upload_file.filename}") try: # Use asyncio to read the file chunks in a non-blocking manner async with aiofiles.open(temp_filename, 'wb') as out_file: while content := await upload_file.read(1024 * 1024): await out_file.write(content) return temp_filename except Exception as e: logger.error(f"Error saving file: {e}") raise HTTPException(status_code=500, detail="Could not save reference audio file") # --- FastAPI Lifespan Manager (Kokoro Feature) --- @asynccontextmanager async def lifespan(app: FastAPI): """Modern lifespan management: initialize model on startup, shutdown executor.""" try: app.state.tts_wrapper = NeuTTSWrapper(device=DEVICE) except Exception as e: logger.error(f"Fatal startup error: {e}") # Terminate the application if the model can't load tts_executor.shutdown(wait=False) raise RuntimeError("Model initialization failed.") yield # Application serves requests # Shutdown logger.info("Shutting down ThreadPoolExecutor.") tts_executor.shutdown(wait=False) # --- FastAPI Application Setup --- app = FastAPI( title="NeuTTS Air Instant Cloning API", version="2.0.0-PROD-ENHANCED", docs_url="/docs", lifespan=lifespan ) app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_methods=["*"], allow_headers=["*"], ) # --- New Endpoints and Enhancements --- @app.get("/") async def root(): return {"message": "NeuTTS Air API v2.0 - Ready for Instant Voice Cloning"} @app.get("/health") async def health_check(): """Enhanced health check (Kokoro Feature + Original Metrics)""" mem = psutil.virtual_memory() disk = psutil.disk_usage('/') return { "status": "healthy", "model_loaded": hasattr(app.state, 'tts_wrapper') and app.state.tts_wrapper.tts_model is not None, "device": DEVICE, "concurrency_limit": MAX_WORKERS, "memory_usage": { "total_gb": round(mem.total / (1024**3), 2), "used_percent": mem.percent }, "disk_usage": { "total_gb": round(disk.total / (1024**3), 2), "used_percent": disk.percent } } @app.delete("/cleanup") async def cleanup_files(): """Maintenance endpoint to remove old generated and temporary files.""" await run_blocking_task_async(cleanup_files_blocking) return {"message": "Cleanup initiated successfully."} def cleanup_files_blocking(): """Blocking file cleanup logic (original NeuTTS feature).""" now = time.time() deleted_count = 0 for directory in [GENERATED_AUDIO_DIR, TEMP_AUDIO_DIR]: for filename in os.listdir(directory): filepath = os.path.join(directory, filename) if os.path.isfile(filepath): try: # Original cleanup logic: delete if older than CLEANUP_THRESHOLD if now - os.path.getctime(filepath) > CLEANUP_THRESHOLD: os.remove(filepath) deleted_count += 1 except Exception as e: logger.warning(f"Failed to delete {filepath}: {e}") logger.info(f"Cleanup completed: {deleted_count} files removed.") return deleted_count # --- Core Synthesis Endpoints --- @app.post("/synthesize", response_class=Response) async def text_to_speech( text: str = Form(...), reference_text: str = Form(...), speed: float = Form(1.0, ge=0.5, le=2.0), output_format: str = Form("wav", pattern="^(wav|mp3|flac)$"), reference_audio: UploadFile = File(...)): """ Standard blocking TTS endpoint with in-memory processing and caching. """ if not hasattr(app.state, 'tts_wrapper'): raise HTTPException(status_code=503, detail="Service unavailable: Model not loaded") start_time = time.time() try: # 1. Convert the uploaded file to WAV directly in memory converted_wav_buffer = await convert_to_wav_in_memory(reference_audio) ref_audio_bytes = converted_wav_buffer.getvalue() # 2. Offload the blocking AI process (now faster with caching) audio_data = await run_blocking_task_async( app.state.tts_wrapper.generate_speech_blocking, text, ref_audio_bytes, # Pass bytes, not a path reference_text ) # 3. Convert to requested output format audio_bytes = await run_blocking_task_async( app.state.tts_wrapper._convert_to_streamable_format, audio_data, output_format ) processing_time = time.time() - start_time audio_duration = len(audio_data) / SAMPLE_RATE return Response( content=audio_bytes, media_type=f"audio/{'mpeg' if output_format == 'mp3' else output_format}", headers={ "Content-Disposition": f"attachment; filename=tts_output.{output_format}", "X-Processing-Time": f"{processing_time:.2f}s", "X-Audio-Duration": f"{audio_duration:.2f}s" } ) except Exception as e: logger.error(f"Synthesis error: {e}") if isinstance(e, HTTPException): raise raise HTTPException(status_code=500, detail=f"Synthesis failed: {e}") @app.post("/synthesize/stream") async def stream_text_to_speech_cloning( text: str = Form(..., min_length=1, max_length=5000), reference_text: str = Form(...), speed: float = Form(1.0, ge=0.5, le=2.0), output_format: str = Form("mp3", pattern="^(wav|mp3|flac)$"), reference_audio: UploadFile = File(...)): """ Sentence-by-Sentence Streaming using a high-performance, asyncio-native producer-consumer pipeline. This overlaps CPU-bound AI work with network I/O. """ if not hasattr(app.state, 'tts_wrapper'): raise HTTPException(status_code=503, detail="Service unavailable: Model not loaded") # This async generator is the final, correct implementation. async def stream_generator(): loop = asyncio.get_event_loop() q = asyncio.Queue(maxsize=2) # The PRODUCER is now an async task that runs in the background. async def producer(): try: # The one-time setup cost: convert and encode the reference voice. # This is done before the loop to ensure the voice is ready. converted_wav_buffer = await convert_to_wav_in_memory(reference_audio) ref_audio_bytes = converted_wav_buffer.getvalue() audio_hash = hashlib.sha256(ref_audio_bytes).hexdigest() ref_s = await loop.run_in_executor( tts_executor, app.state.tts_wrapper._get_or_create_reference_encoding, audio_hash, ref_audio_bytes ) sentences = app.state.tts_wrapper._split_text_into_chunks(text) for sentence in sentences: # Define the blocking work for a single chunk def process_chunk(): with torch.no_grad(): audio_chunk = app.state.tts_wrapper.tts_model.infer(sentence, ref_s, reference_text) return app.state.tts_wrapper._convert_to_streamable_format(audio_chunk, output_format) # Offload the blocking work to the thread pool mp3_bytes = await loop.run_in_executor(tts_executor, process_chunk) # Put the finished MP3 chunk into the async queue await q.put(mp3_bytes) except Exception as e: logger.error(f"Error in producer task: {e}") await q.put(e) finally: # Signal that production is finished await q.put(None) # Start the producer as a background task. It starts working immediately. producer_task = asyncio.create_task(producer()) # The main loop now acts as the CONSUMER. while True: # Await the next finished MP3 chunk from the queue. result = await q.get() if result is None: break if isinstance(result, Exception): logger.error(f"Terminating stream due to producer error: {result}") raise result # Yield the chunk to the user. While the network sends this, # the producer is already working on the next chunk in the background. yield result # Ensure the producer task is cleaned up. await producer_task return StreamingResponse( stream_generator(), media_type=f"audio/{'mpeg' if output_format == 'mp3' else output_format}" ) # Note: The outer 'finally' block is now removed as its logic is handled in 2.5 and 4. @app.get("/audio/{filename}") async def get_audio(filename: str): """Original NeuTTS feature to serve generated audio files.""" file_path = os.path.join(GENERATED_AUDIO_DIR, filename) if not os.path.exists(file_path): raise HTTPException(status_code=404, detail="Audio file not found") return Response( content=open(file_path, "rb").read(), media_type=f"audio/{filename.split('.')[-1]}", # Simple media type detection headers={"Content-Disposition": f"attachment; filename={filename}"} )