server.py

from onnx_modules.V230_OnnxInference import OnnxInferenceSession
import numpy as np
import torch
from scipy.io.wavfile import write
from text import cleaned_text_to_sequence, get_bert
from text.cleaner import clean_text
import utils
import commons
import uuid
from flask import Flask, request, jsonify, render_template_string
from flask_cors import CORS
import gradio as gr
import os
from threading import Thread

hps = utils.get_hparams_from_file('onnx/BangDreamApi.json')
device = 'cpu'

BandList = {
        "PoppinParty":["香澄","有咲","たえ","りみ","沙綾"],
        "Afterglow":["蘭","モカ","ひまり","巴","つぐみ"],
        "HelloHappyWorld":["こころ","美咲","薫","花音","はぐみ"],
        "PastelPalettes":["彩","日菜","千聖","イヴ","麻弥"],
        "Roselia":["友希那","紗夜","リサ","燐子","あこ"],
        "RaiseASuilen":["レイヤ","ロック","ますき","チュチュ","パレオ"],
        "Morfonica":["ましろ","瑠唯","つくし","七深","透子"],
        "MyGo":["燈","愛音","そよ","立希","楽奈"],
        "AveMujica":["祥子","睦","海鈴","にゃむ","初華"],
        "圣翔音乐学园":["華戀","光","香子","雙葉","真晝","純那","克洛迪娜","真矢","奈奈"],
        "凛明馆女子学校":["珠緒","壘","文","悠悠子","一愛"],
        "弗隆提亚艺术学校":["艾露","艾露露","菈樂菲","司","靜羽"],
        "西克菲尔特音乐学院":["晶","未知留","八千代","栞","美帆"]
}

Session = OnnxInferenceSession(
        {
        "enc" : "onnx/BangDreamApi/BangDreamApi_enc_p.onnx",
        "emb_g" : "onnx/BangDreamApi/BangDreamApi_emb.onnx",
        "dp" : "onnx/BangDreamApi/BangDreamApi_dp.onnx",
        "sdp" : "onnx/BangDreamApi/BangDreamApi_sdp.onnx",
        "flow" : "onnx/BangDreamApi/BangDreamApi_flow.onnx",
        "dec" : "onnx/BangDreamApi/BangDreamApi_dec.onnx"
        },
        Providers = ["CPUExecutionProvider"]
    )

def get_text(text, language_str, hps, device, style_text=None, style_weight=0.7):
    style_text = None if style_text == "" else style_text
    norm_text, phone, tone, word2ph = clean_text(text, language_str)
    phone, tone, language = cleaned_text_to_sequence(phone, tone, language_str)

    if True:
        phone = commons.intersperse(phone, 0)
        tone = commons.intersperse(tone, 0)
        language = commons.intersperse(language, 0)
        for i in range(len(word2ph)):
            word2ph[i] = word2ph[i] * 2
        word2ph[0] += 1
    bert_ori = get_bert(
        norm_text, word2ph, language_str, device, style_text, style_weight
    )
    del word2ph
    assert bert_ori.shape[-1] == len(phone), phone

    if language_str == "ZH":
        bert = bert_ori
        ja_bert = torch.randn(1024, len(phone))
        en_bert = torch.randn(1024, len(phone))
    elif language_str == "JP":
        bert = torch.randn(1024, len(phone))
        ja_bert = bert_ori
        en_bert = torch.randn(1024, len(phone))
    elif language_str == "EN":
        bert = torch.randn(1024, len(phone))
        ja_bert = torch.randn(1024, len(phone))
        en_bert = bert_ori
    else:
        raise ValueError("language_str should be ZH, JP or EN")

    assert bert.shape[-1] == len(
        phone
    ), f"Bert seq len {bert.shape[-1]} != {len(phone)}"

    phone = torch.LongTensor(phone)
    tone = torch.LongTensor(tone)
    language = torch.LongTensor(language)
    return bert, ja_bert, en_bert, phone, tone, language


def infer(
    text,
    sid,
    style_text=None,
    style_weight=0.7,
    sdp_ratio=0.5,
    noise_scale=0.6,
    noise_scale_w=0.667,
    length_scale=1,
    unique_filename = 'temp.wav'
):
    language= 'JP' if is_japanese(text) else 'ZH'
    bert, ja_bert, en_bert, phones, tone, language = get_text(
        text,
        language,
        hps, 
        device,
        style_text=style_text,
        style_weight=style_weight,
    )
    with torch.no_grad():
        x_tst = phones.unsqueeze(0).to(device).numpy()
        language = np.zeros_like(x_tst)
        tone = np.zeros_like(x_tst)
        bert = bert.to(device).transpose(0, 1).numpy()
        ja_bert = ja_bert.to(device).transpose(0, 1).numpy()
        en_bert = en_bert.to(device).transpose(0, 1).numpy()
        del phones
        sid = np.array([hps.spk2id[sid]])
        audio = Session(
                x_tst,
                tone,
                language,
                bert,
                ja_bert,
                en_bert,
                sid,
                seed=114514,
                seq_noise_scale=noise_scale_w,
                sdp_noise_scale=noise_scale,
                length_scale=length_scale,
                sdp_ratio=sdp_ratio,
            )
        del x_tst, tone, language, bert, ja_bert, en_bert, sid
        write(unique_filename, 44100, audio)
        return (44100,gr.processing_utils.convert_to_16_bit_wav(audio))

def is_japanese(string):
        for ch in string:
            if ord(ch) > 0x3040 and ord(ch) < 0x30FF:
                return True
        return False


Flaskapp = Flask(__name__)
CORS(Flaskapp)
@Flaskapp.route('/')

def tts():
    global last_text, last_model
    speaker = request.args.get('speaker')
    sdp_ratio = float(request.args.get('sdp_ratio', 0.2))
    noise_scale = float(request.args.get('noise_scale', 0.6))
    noise_scale_w = float(request.args.get('noise_scale_w', 0.8))
    length_scale = float(request.args.get('length_scale', 1))
    style_weight = float(request.args.get('style_weight', 0.7))
    style_text = request.args.get('style_text', 'happy')
    text = request.args.get('text')
    is_chat = request.args.get('is_chat', 'false').lower() == 'true'
    #model = request.args.get('model',modelPaths[-1])
    
    if not speaker or not text:
        return render_template_string("""
            <!DOCTYPE html>
            <html>
            <head>
                <title>TTS API Documentation</title>
            </head>
            <body>
                <iframe src="https://mahiruoshi-bangdream-bert-vits2.hf.space" style="width:100%; height:100vh; border:none;"></iframe>
            </body>
            </html>
        """)
    '''
    if model != last_model:
        unique_filename  = loadmodel(model)
        last_model = model
    '''
    if is_chat and text == last_text:
        # Generate 1 second of silence and return
        unique_filename = 'blank.wav'
        silence = np.zeros(44100, dtype=np.int16)
        write(unique_filename , 44100, silence)
    else:
        last_text = text
        unique_filename = f"temp{uuid.uuid4()}.wav"
        infer(text, sdp_ratio=sdp_ratio, noise_scale=noise_scale, noise_scale_w=noise_scale_w, length_scale=length_scale,sid = speaker, style_text=style_text, style_weight=style_weight,unique_filename=unique_filename)
    with open(unique_filename ,'rb') as bit:
        wav_bytes = bit.read()
    os.remove(unique_filename)
    headers = {
            'Content-Type': 'audio/wav',
            'Text': unique_filename .encode('utf-8')}
    return wav_bytes, 200, headers

if __name__ == "__main__":
    speaker_ids = hps.spk2id
    speakers = list(speaker_ids.keys())
    last_text = ""
    Flaskapp.run(host="0.0.0.0", port=5000,debug=True)