dia16b/app/diaapp.py
2025-05-11 06:30:45 +00:00

66 lines
1.9 KiB
Python

import os
import asyncio
import soundfile as sf
import numpy as np
from traceback import format_exc
from dia.model import Dia
from appPublic.worker import awaitify
from appPublic.hf import hf_socks5proxy
from appPublic.folderUtils import _mkdir
from appPublic.uniqueID import getID
from appPublic.log import debug, exception, error
from appPublic.jsonConfig import getConfig
from ahserver.filestorage import FileStorage
from ahserver.webapp import webapp
from ahserver.serverenv import ServerEnv
import dac
hf_socks5proxy()
class DiaTTS:
def __init__(self):
config = getConfig()
# self.model = Dia.from_local(config.dia_model_path, compute_dtype="float16")
self.model = Dia.from_local(config.dia_model_path+'/config.json',
device=config.device or 'cpu',
load_dac=False,
checkpoint_path=config.dia_model_path + '/dia-v0_1.pth',
compute_dtype="float16")
self.load_dac(config.dac_model_path)
self.lock = asyncio.Lock()
self.fs = FileStorage()
def load_dac(self, dac_model_path):
dac_model = dac.DAC.load(dac_model_path).to(self.model.device)
dac_model.eval() # Ensure DAC is in eval mode
self.model.dac_model = dac_model
def _generate(self, prompt):
name = getID() + '.wav'
fp = self.fs._name2path(name)
webpath = self.fs.webpath(fp)
debug(f'{prompt=}')
output = self.model.generate(prompt, use_torch_compile=True, verbose=True)
output = output.astype(np.float32)
if output is None:
e = Exception(f'"{prompt}" to audio null')
exception(f'{e}\n{format_exc()}')
raise e
debug(f'{output.dtype.name=}')
_mkdir(os.path.dirname(fp))
sf.write(fp, output, 44100)
return webpath
async def generate(self, prompt):
async with self.lock:
f = awaitify(self._generate)
return await f(prompt)
def init():
g = ServerEnv()
g.etts_engine = DiaTTS()
if __name__ == '__main__':
webapp(init)