bench.py

import requests
import random
import json
from time import perf_counter
import argparse


def synthesis(
    text: str, address="127.0.0.1", port=50021, speaker=0, pitch=0.0, speed=1.0
) -> float:
    """音声生成

    Args:
        text : 生成する文章
        address : VOCIEVOXのAPIサーバーのアドレス
        speaker : speaker_id
        pitch : ピッチ
        speed : スピード

    Returns:
        生成時間(秒)
    """
    address = "http://"+address
    query_payload = {"text": text, "speaker": speaker}
    resp = requests.post(f"{address}:{port}/audio_query", params=query_payload)
    if not resp.status_code == 200:
        raise ConnectionError("Status code: %d" % resp.status_code)
    query_data = resp.json()
    synth_payload = {"speaker": speaker}
    query_data["speedScale"] = speed
    query_data["pitchScale"] = pitch
    before = perf_counter()
    resp = requests.post(f"{address}:{port}/synthesis",
                         params=synth_payload, data=json.dumps(query_data))
    if not resp.status_code == 200:
        raise ConnectionError("Status code: %d" % resp.status_code)
    after = perf_counter()
    return after - before


def gen_text(count: int):
    return "".join(
        [chr(random.randint(ord("あ"), ord("ん"))) for i in range(count)])


def get_speakers(address="127.0.0.1", port=50021):
    address = "http://"+address
    speakers = {}
    resp = requests.get(f"{address}:{port}/speakers")
    resp_dict = resp.json()
    for i in resp_dict:
        speakers[i["name"]] = {}
        for s in i["styles"]:
            speakers[i["name"]][s["name"]] = s["id"]
    return speakers


def bench(length: int, count=10, address="127.0.0.1", port=50021, quiet=False):
    synthesis("test", address=address, port=port)
    tmp = 0
    for i in range(count):
        text = gen_text(length)
        elapsed_time = synthesis(text, address=address, port=port)
        tmp += elapsed_time
        if not quiet:
            print(i+1, "time:", elapsed_time)
        result = round(tmp / count, 4)
    return result


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "-s", help="VOICEVOX API Server Address", default="127.0.0.1")
    parser.add_argument("-p", help="VOICEVOX API Server Port", default=50021)
    parser.add_argument("-q", help="Quiet benchmark log", action="store_true")
    parser.add_argument("-w", help="No wait for key input",
                        action="store_true")
    parser.add_argument("--ssl", help="Use SSL", action="store_true")
    args = parser.parse_args()
    if not args.w:
        input("Press Enter key to start benchmark...")
    if args.ssl:
        ADDRESS = "https://" + args.s
    else:
        ADDRESS = "http://" + args.s
    score_10 = bench(length=10, address=args.s, port=args.p, quiet=args.q)
    score_50 = bench(length=50, address=args.s, port=args.p, quiet=args.q)
    score_100 = bench(length=100, address=args.s, port=args.p, quiet=args.q)
    score_avg = round((score_10 + score_50 + score_100) / 3, 4)
    resp = requests.get(f"{ADDRESS}:{args.p}/version")
    info_engine = resp.text.replace("\"", "")
    resp = requests.get(f"{ADDRESS}:{args.p}/supported_devices")
    info_devices = resp.json()
    if info_devices["cuda"]:
        info_device = "CUDA"
    elif info_devices["dml"]:
        info_device = "DirectML"
    else:
        info_device = "CPU"
    print()
    print("=========== Info ===========")
    print(" Engine:", info_engine)
    print(" Device:", info_device)
    print("========== Result ==========")
    print(" 10: ", score_10)
    print(" 50: ", score_50)
    print(" 100:", score_100)
    print(" Avg:", score_avg)
    print("============================")
    print()
    if not args.w:
        input("Press Enter key to exit...")