main.py

from fastapi import FastAPI, HTTPException, Query
from fastapi.responses import StreamingResponse
import os
from os import environ as env
import torch
import time
import nltk
import io
import base64
import torchaudio
from fastapi.responses import JSONResponse
from app.inference import inference, LFinference, compute_style
import numpy as np

nltk.download('punkt')
nltk.download('punkt_tab')

app = FastAPI()

device = 'cuda' if torch.cuda.is_available() else 'cpu'

@app.get("/")
async def read_root():
    #return {"details": f"Hello! This is {env['SECRET_API_KEY']} environment"}
    #return {"details": f"Hello Stream!"}
    #return {"details": f"Hello Stream! This is {env['API_KEY_SECRET']} environment running OK!"}
    return {"details": "Environment is running OK!"}

@app.post("/synthesize/")
async def synthesize(
    text: str, 
    return_base64: bool = True,
    ###################################################
    diffusion_steps: int = Query(5, ge=5, le=200),
    embedding_scale: float = Query(1.0, ge=1.0, le=5.0)
    ###################################################
):
    try:
        start = time.time()
        noise = torch.randn(1, 1, 256).to(device)
        wav = inference(text, noise, diffusion_steps=diffusion_steps, embedding_scale=embedding_scale)
        rtf = (time.time() - start) / (len(wav) / 24000)

        if return_base64:
            audio_buffer = io.BytesIO()
            torchaudio.save(audio_buffer, torch.tensor(wav).unsqueeze(0), 24000, format="wav")
            audio_buffer.seek(0)

            audio_base64 = base64.b64encode(audio_buffer.read()).decode('utf-8')

            return JSONResponse(content={"RTF": rtf, "audio_base64": audio_base64})
        else:
            return JSONResponse(content={"RTF": rtf, "audio": wav.tolist()})

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))


@app.post("/synthesize_longform_streaming/")
async def synthesize_longform(
    passage: str, 
    return_base64: bool = False,
    ###################################################
    alpha: float = Query(0.7, ge=0.0, le=1.0),
    diffusion_steps: int = Query(10, ge=5, le=200),
    embedding_scale: float = Query(1.5, ge=1.0, le=5.0)
    ###################################################
):
    try:
        sentences = passage.split('.') # simple split
        wavs = []
        s_prev = None
        
        start = time.time()
        
        for text in sentences:
            if text.strip() == "":
                continue
            text += '.'  # add it back
            noise = torch.randn(1, 1, 256).to(device)  # Generate noise
            wav, s_prev = LFinference(text, s_prev, noise, alpha=0.7, 
                                       diffusion_steps=diffusion_steps, 
                                       embedding_scale=embedding_scale)
            wavs.append(wav)

        final_wav = np.concatenate(wavs)  # Concatenate all wavs
        rtf = (time.time() - start) / (len(final_wav) / 24000)

        audio_buffer = io.BytesIO()
        torchaudio.save(audio_buffer, torch.tensor(final_wav).unsqueeze(0), 24000, format="wav")
        audio_buffer.seek(0)

        if return_base64:
            audio_base64 = base64.b64encode(audio_buffer.read()).decode('utf-8')
            return JSONResponse(content={"RTF": rtf, "audio_base64": audio_base64})
        else:
            #return JSONResponse(content={"RTF": rtf, "audio": final_wav.tolist()})
            return StreamingResponse(audio_buffer, media_type="audio/wav")

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))



@app.post("/synthesize_with_emotion/")
async def synthesize_with_emotion(
    texts: dict, 
    return_base64: bool = True,
    ###################################################
    diffusion_steps: int = Query(100, ge=5, le=200),
    embedding_scale: float = Query(5.0, ge=1.0, le=5.0)
    ###################################################
):
    try:
        results = []
        
        for emotion, text in texts.items():
            noise = torch.randn(1, 1, 256).to(device)
            wav = inference(text, noise, diffusion_steps=diffusion_steps, 
                            embedding_scale=embedding_scale)
            
            if return_base64:
                audio_buffer = io.BytesIO()
                torchaudio.save(audio_buffer, torch.tensor(wav).unsqueeze(0), 24000, format="wav")
                audio_buffer.seek(0)

                audio_base64 = base64.b64encode(audio_buffer.read()).decode('utf-8')

                results.append({
                    "emotion": emotion,
                    "audio_base64": audio_base64
                })
            else:
                results.append({
                    "emotion": emotion,
                    "audio": wav.tolist()
                })

        return JSONResponse(content={"results": results})

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))





@app.post("/synthesize_streaming_audio/")
async def synthesize_streaming_audio(
    text: str, 
    return_base64: bool = False,
    ###################################################
    diffusion_steps: int = Query(5, ge=5, le=200),
    embedding_scale: float = Query(1.0, ge=1.0, le=5.0)
    ###################################################
):
    try:
        start = time.time()
        noise = torch.randn(1, 1, 256).to(device)
        wav = inference(text, noise, diffusion_steps=diffusion_steps, embedding_scale=embedding_scale)
        rtf = (time.time() - start) / (len(wav) / 24000)

        audio_buffer = io.BytesIO()
        torchaudio.save(audio_buffer, torch.tensor(wav).unsqueeze(0), 24000, format="wav")
        audio_buffer.seek(0)

        if return_base64:
            audio_base64 = base64.b64encode(audio_buffer.read()).decode('utf-8')
            return JSONResponse(content={"RTF": rtf, "audio_base64": audio_base64})
        else:
            return StreamingResponse(audio_buffer, media_type="audio/wav")

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))