main.py

from typing import Union
from fastapi import FastAPI
from fastapi.responses import FileResponse
from pydantic import BaseModel
from datetime import datetime
import re
from balacoon_tts import TTS
import wave

# Model can be dowloaded in huggingface
# from huggingface_hub import hf_hub_download, list_repo_files
# model_path = hf_hub_download(repo_id="balacoon/tts", filename="en_us_hifi92_light_cpu.addon")
model_path = "model/en_us_hifi92_light_cpu.addon"
tts = TTS(model_path)
speakers = tts.get_speakers()

app = FastAPI(
    title="REST_TTS_Dockerized",
    description="TTS REST API wrapper with FastAPI and Docker, for your text-to-speech needs🤖",
    summary="TTS REST API Dockerized",
    version="0.0.1",
    terms_of_service="https://github.com/rsandagon/REST_TTS_Dockerized/README.md",
    contact={
        "name": "rsandagon",
        "url": "https://github.com/rsandagon",
        "email": "rsandagon.com",
    },
    license_info={
        "name": "Apache 2.0",
        "url": "https://www.apache.org/licenses/LICENSE-2.0.html",
    },
)

class Item(BaseModel):
    name: str
    message: str
    sample_rate: float 

def remove_between_asterisks(text):
  pattern = r"\*.*?\*"
  return re.sub(pattern, "", text)

@app.get("/")
def read_root():
    return {"Hello": "I'm alive!"}


@app.get(
    path="/api/snd"
)
async def post_media_file(name:str='default'):
    return FileResponse("audio/outputs/"+name, media_type="audio/mpeg")

@app.post("/api/tts")
async def post_tts(item: Item):
    now = datetime.now()
    nowstr = now.strftime("%m%d%Y_%H%M%S")
    outname = item.name+nowstr+".wav"
    sample_rate = item.sample_rate

    samples = tts.synthesize(remove_between_asterisks(item.message), speakers[-1])

    with wave.open("audio/outputs/"+outname, "w") as fp:
        fp.setparams((1, 2, tts.get_sampling_rate()*sample_rate, len(samples), "NONE", "NONE"))
        fp.writeframes(samples)

    return {"file":outname}