Ma/auto-caption
1
0
Fork 0
mirror of https://github.com/HiMeditator/auto-caption.git synced 2026-04-11 23:37:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor(engine): 重构字幕引擎

Browse Source 
- 更新 GummyTranslator 类，优化字幕生成逻辑
- 移除 audioprcs 模块，音频处理功能转移到 utils 模块
- 重构 sysaudio 模块，提高音频流管理的灵活性和稳定性
- 修改 TODO.md，完成按时间降序排列字幕记录的功能
- 更新文档，说明因资源限制将不再维护英文和日文文档
This commit is contained in:
himeditator
2025-07-26 23:37:24 +08:00
parent 8e575a9ba3
commit 3792eb88b6
18 changed files with 245 additions and 136 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.gitignore vendored
View File

@@ -9,4 +9,4 @@ __pycache__
subenv
engine/build
engine/models
engine/test
engine/notebook

2
docs/TODO.md
View File

@@ -15,10 +15,10 @@
- [x] 可以调整字幕时间轴 *2025/07/14*
- [x] 可以导出 srt 格式的字幕记录 *2025/07/14*
- [x] 可以获取字幕引擎的系统资源消耗情况 *2025/07/15*
- [x] 添加字幕记录按时间降序排列选择 *2025/07/26*
## 待完成
- [ ] 修改字幕记录展示逻辑
- [ ] 重构字幕引擎
- [ ] 验证 / 添加基于 sherpa-onnx 的字幕引擎

51
docs/api-docs/caption-engine.md Normal file
View File

@@ -0,0 +1,51 @@
# caption engine api-doc
本文档主要 Electron 主进程和字幕引擎进程的通信约定。
## 原理说明
本项目的 Python 进程通过标准输出向 Electron 主进程发送数据。
Python 进程标准输出 (`sys.stdout`) 的内容一定为一行一行的字符串。且每行字符串均可以解释为一个 JSON 对象。每个 JSON 对象一定有 `command` 参数。
## 输出约定
当 JSON 对象的 `command` 参数为下列值时，表示的对应的含义：
### `print`
```js
{
command: "print",
content: string
}
```
输出 Python 端打印的内容。
### `info`
```js
{
command: "info",
content: string
}
```
Python 端打印的提示信息。
### `caption`
```js
{
command: "caption",
index: number,
time_s: string,
time_t: string,
end: boolean,
text: string,
translation: string
}
```
Python 端监听到的音频流转换为的字幕数据。

2
docs/engine-manual/en.md
View File

@@ -2,6 +2,8 @@
Corresponding Version: v0.5.1
**Note: Due to limited personal resources, the English and Japanese documentation files for this project (except for the README document) will no longer be maintained. The content of this document may not be consistent with the latest version of the project. If you are willing to help with translation, please submit relevant Pull Requests.**
![](../../assets/media/structure_en.png)
## Introduction to the Caption Engine

2
docs/engine-manual/ja.md
View File

@@ -4,6 +4,8 @@
この文書は大規模モデルを使用して翻訳されていますので、内容に正確でない部分があるかもしれません。
**注意：個人のリソースが限られているため、このプロジェクトの英語および日本語のドキュメント（README ドキュメントを除く）のメンテナンスは行われません。このドキュメントの内容は最新版のプロジェクトと一致しない場合があります。翻訳のお手伝いをしていただける場合は、関連するプルリクエストを提出してください。**
![](../../assets/media/structure_ja.png)
## 字幕エンジンの紹介

2
docs/user-manual/en.md
View File

@@ -2,6 +2,8 @@
Corresponding Version: v0.5.1
**Note: Due to limited personal resources, the English and Japanese documentation files for this project (except for the README document) will no longer be maintained. The content of this document may not be consistent with the latest version of the project. If you are willing to help with translation, please submit relevant Pull Requests.**
## Software Introduction
Auto Caption is a cross-platform caption display software that can real-time capture system audio input (recording) or output (playback) streaming data and use an audio-to-text model to generate captions for the corresponding audio. The default caption engine provided by the software (using Alibaba Cloud Gummy model) supports recognition and translation in nine languages (Chinese, English, Japanese, Korean, German, French, Russian, Spanish, Italian).

2
docs/user-manual/ja.md
View File

@@ -4,6 +4,8 @@
この文書は大規模モデルを使用して翻訳されていますので、内容に正確でない部分があるかもしれません。
**注意：個人のリソースが限られているため、このプロジェクトの英語および日本語のドキュメント（README ドキュメントを除く）のメンテナンスは行われません。このドキュメントの内容は最新版のプロジェクトと一致しない場合があります。翻訳のお手伝いをしていただける場合は、関連するプルリクエストを提出してください。**
## ソフトウェアの概要
Auto Caption は、クロスプラットフォームの字幕表示ソフトウェアで、システムの音声入力（録音）または出力（音声再生）のストリーミングデータをリアルタイムで取得し、音声からテキストに変換するモデルを利用して対応する音声の字幕を生成します。このソフトウェアが提供するデフォルトの字幕エンジン（アリババクラウド Gummy モデルを使用）は、9つの言語（中国語、英語、日本語、韓国語、ドイツ語、フランス語、ロシア語、スペイン語、イタリア語）の認識と翻訳をサポートしています。

47
engine/audio2text/gummy.py
View File

@@ -6,8 +6,8 @@ from dashscope.audio.asr import (
)
import dashscope
from datetime import datetime
import json
import sys
from utils import stdout_cmd, stdout_obj
class Callback(TranslationRecognizerCallback):
"""
@@ -17,15 +17,16 @@ class Callback(TranslationRecognizerCallback):
super().__init__()
self.usage = 0
self.cur_id = -1
self.index = 0
self.time_str = ''
def on_open(self) -> None:
# print("on_open")
pass
self.cur_id = -1
self.time_str = ''
stdout_cmd('info', 'Gummy translator started.')
def on_close(self) -> None:
# print("on_close")
pass
stdout_cmd('info', 'Gummy translator closed.')
def on_event(
self,
@@ -35,17 +36,18 @@ class Callback(TranslationRecognizerCallback):
usage
) -> None:
caption = {}
if transcription_result is not None:
caption['index'] = transcription_result.sentence_id
caption['text'] = transcription_result.text
if caption['index'] != self.cur_id:
self.cur_id = caption['index']
cur_time = datetime.now().strftime('%H:%M:%S.%f')[:-3]
caption['time_s'] = cur_time
self.time_str = cur_time
else:
caption['time_s'] = self.time_str
if self.cur_id != transcription_result.sentence_id:
self.time_str = datetime.now().strftime('%H:%M:%S.%f')[:-3]
self.cur_id = transcription_result.sentence_id
self.index += 1
caption['command'] = 'caption'
caption['index'] = self.index
caption['time_s'] = self.time_str
caption['time_t'] = datetime.now().strftime('%H:%M:%S.%f')[:-3]
caption['end'] = transcription_result.is_sentence_end
caption['text'] = transcription_result.text
caption['translation'] = ""
if translation_result is not None:
@@ -55,19 +57,8 @@ class Callback(TranslationRecognizerCallback):
if usage:
self.usage += usage['duration']
# print(caption)
self.send_to_node(caption)
stdout_obj(caption)
def send_to_node(self, data):
"""
将数据发送到 Node.js 进程
"""
try:
json_data = json.dumps(data) + '\n'
sys.stdout.write(json_data)
sys.stdout.flush()
except Exception as e:
print(f"Error sending data to Node.js: {e}", file=sys.stderr)
class GummyTranslator:
"""
@@ -78,7 +69,7 @@ class GummyTranslator:
source: 源语言代码字符串（zh, en, ja 等）
target: 目标语言代码字符串（zh, en, ja 等）
"""
def __init__(self, rate, source, target, api_key):
def __init__(self, rate: int, source: str, target: str | None, api_key: str | None):
if api_key:
dashscope.api_key = api_key
self.translator = TranslationRecognizerRealtime(

1
engine/audioprcs/__init__.py
View File

@@ -1 +0,0 @@
from .process import mergeChunkChannels, resampleRawChunk, resampleMonoChunk

21
engine/main-gummy.py
View File

@@ -1,21 +1,11 @@
import sys
import argparse
if sys.platform == 'win32':
from sysaudio.win import AudioStream
elif sys.platform == 'darwin':
from sysaudio.darwin import AudioStream
elif sys.platform == 'linux':
from sysaudio.linux import AudioStream
else:
raise NotImplementedError(f"Unsupported platform: {sys.platform}")
from audioprcs import mergeChunkChannels
from sysaudio import AudioStream
from utils import merge_chunk_channels
from audio2text import InvalidParameter, GummyTranslator
def convert_audio_to_text(s_lang, t_lang, audio_type, chunk_rate, api_key):
sys.stdout.reconfigure(line_buffering=True) # type: ignore
stream = AudioStream(audio_type, chunk_rate)
if t_lang == 'none':
@@ -23,20 +13,21 @@ def convert_audio_to_text(s_lang, t_lang, audio_type, chunk_rate, api_key):
else:
gummy = GummyTranslator(stream.RATE, s_lang, t_lang, api_key)
stream.openStream()
stream.open_stream()
gummy.start()
while True:
try:
chunk = stream.read_chunk()
chunk_mono = mergeChunkChannels(chunk, stream.CHANNELS)
if chunk is None: continue
chunk_mono = merge_chunk_channels(chunk, stream.CHANNELS)
try:
gummy.send_audio_frame(chunk_mono)
except InvalidParameter:
gummy.start()
gummy.send_audio_frame(chunk_mono)
except KeyboardInterrupt:
stream.closeStream()
stream.close_stream()
gummy.stop()
break

17
engine/main-vosk.py
View File

@@ -4,17 +4,9 @@ import argparse
from datetime import datetime
import numpy.core.multiarray
if sys.platform == 'win32':
from sysaudio.win import AudioStream
elif sys.platform == 'darwin':
from sysaudio.darwin import AudioStream
elif sys.platform == 'linux':
from sysaudio.linux import AudioStream
else:
raise NotImplementedError(f"Unsupported platform: {sys.platform}")
from sysaudio import AudioStream
from vosk import Model, KaldiRecognizer, SetLogLevel
from audioprcs import resampleRawChunk
from utils import resample_chunk_mono
SetLogLevel(-1)
@@ -30,7 +22,7 @@ def convert_audio_to_text(audio_type, chunk_rate, model_path):
recognizer = KaldiRecognizer(model, 16000)
stream = AudioStream(audio_type, chunk_rate)
stream.openStream()
stream.open_stream()
time_str = ''
cur_id = 0
@@ -38,7 +30,8 @@ def convert_audio_to_text(audio_type, chunk_rate, model_path):
while True:
chunk = stream.read_chunk()
chunk_mono = resampleRawChunk(chunk, stream.CHANNELS, stream.RATE, 16000)
if chunk is None: continue
chunk_mono = resample_chunk_mono(chunk, stream.CHANNELS, stream.RATE, 16000)
caption = {}
if recognizer.AcceptWaveform(chunk_mono):

10
engine/sysaudio/__init__.py
View File

@@ -0,0 +1,10 @@
import sys
if sys.platform == "win32":
from .win import AudioStream
elif sys.platform == "darwin":
from .darwin import AudioStream
elif sys.platform == "linux":
from .linux import AudioStream
else:
raise NotImplementedError(f"Unsupported platform: {sys.platform}")

77
engine/sysaudio/darwin.py
View File

@@ -1,11 +1,24 @@
"""获取 MacOS 系统音频输入/输出流"""
import pyaudio
from textwrap import dedent
def get_blackhole_device(mic: pyaudio.PyAudio):
"""
获取 BlackHole 设备
"""
device_count = mic.get_device_count()
for i in range(device_count):
dev_info = mic.get_device_info_by_index(i)
if 'blackhole' in str(dev_info["name"]).lower():
return dev_info
raise Exception("The device containing BlackHole was not found.")
class AudioStream:
"""
获取系统音频流（支持 BlackHole 作为系统音频输出捕获）
获取系统音频流（如果要捕获输出音频，仅支持 BlackHole 作为系统音频输出捕获）
初始化参数：
audio_type: 0-系统音频输出流（需配合 BlackHole），1-系统音频输入流
@@ -15,46 +28,40 @@ class AudioStream:
self.audio_type = audio_type
self.mic = pyaudio.PyAudio()
if self.audio_type == 0:
self.device = self.getOutputDeviceInfo()
self.device = get_blackhole_device(self.mic)
else:
self.device = self.mic.get_default_input_device_info()
self.stop_signal = False
self.stream = None
self.SAMP_WIDTH = pyaudio.get_sample_size(pyaudio.paInt16)
self.INDEX = self.device["index"]
self.FORMAT = pyaudio.paInt16
self.CHANNELS = self.device["maxInputChannels"]
self.SAMP_WIDTH = pyaudio.get_sample_size(self.FORMAT)
self.CHANNELS = int(self.device["maxInputChannels"])
self.RATE = int(self.device["defaultSampleRate"])
self.CHUNK = self.RATE // chunk_rate
self.INDEX = self.device["index"]
def getOutputDeviceInfo(self):
"""查找指定关键词的输入设备"""
device_count = self.mic.get_device_count()
for i in range(device_count):
dev_info = self.mic.get_device_info_by_index(i)
if 'blackhole' in dev_info["name"].lower():
return dev_info
raise Exception("The device containing BlackHole was not found.")
def printInfo(self):
def get_info(self):
dev_info = f"""
采样输入设备：
采样设备：
- 设备类型：{ "音频输出" if self.audio_type == 0 else "音频输入" }
- 序号：{self.device['index']}
- 名称：{self.device['name']}
- 设备序号：{self.device['index']}
- 设备名称：{self.device['name']}
- 最大输入通道数：{self.device['maxInputChannels']}
- 默认低输入延迟：{self.device['defaultLowInputLatency']}s
- 默认高输入延迟：{self.device['defaultHighInputLatency']}s
- 默认采样率：{self.device['defaultSampleRate']}Hz
- 是否回环设备：{self.device['isLoopbackDevice']}
音频样本块大小：{self.CHUNK}
设备序号：{self.INDEX}
样本格式：{self.FORMAT}
样本位宽：{self.SAMP_WIDTH}
采样格式：{self.FORMAT}
音频通道数：{self.CHANNELS}
音频采样率：{self.RATE}
样本通道数：{self.CHANNELS}
样本采样率：{self.RATE}
样本块大小：{self.CHUNK}
"""
print(dev_info)
return dedent(dev_info).strip()
def openStream(self):
def open_stream(self):
"""
打开并返回系统音频输出流
"""
@@ -72,14 +79,24 @@ class AudioStream:
"""
读取音频数据
"""
if self.stop_signal:
self.close_stream()
return None
if not self.stream: return None
return self.stream.read(self.CHUNK, exception_on_overflow=False)
def closeStream(self):
def close_stream_signal(self):
"""
关闭系统音频输出流
线程安全的关闭系统音频输入流，不一定会立即关闭
"""
if self.stream is None: return
self.stream.stop_stream()
self.stream.close()
self.stream = None
self.stop_signal = True
def close_stream(self):
"""
立即关闭系统音频输入流
"""
if self.stream is not None:
self.stream.stop_stream()
self.stream.close()
self.stream = None
self.stop_signal = False

46
engine/sysaudio/linux.py
View File

@@ -1,8 +1,10 @@
"""获取 Linux 系统音频输入流"""
import subprocess
from textwrap import dedent
def findMonitorSource():
def find_monitor_source():
result = subprocess.run(
["pactl", "list", "short", "sources"],
stdout=subprocess.PIPE, text=True
@@ -16,7 +18,8 @@ def findMonitorSource():
raise RuntimeError("System output monitor device not found")
def findInputSource():
def find_input_source():
result = subprocess.run(
["pactl", "list", "short", "sources"],
stdout=subprocess.PIPE, text=True
@@ -28,8 +31,10 @@ def findInputSource():
name = parts[1]
if ".monitor" not in name:
return name
raise RuntimeError("Microphone input device not found")
class AudioStream:
"""
获取系统音频流
@@ -42,34 +47,33 @@ class AudioStream:
self.audio_type = audio_type
if self.audio_type == 0:
self.source = findMonitorSource()
self.source = find_monitor_source()
else:
self.source = findInputSource()
self.source = find_input_source()
self.stop_signal = False
self.process = None
self.SAMP_WIDTH = 2
self.FORMAT = 16
self.SAMP_WIDTH = 2
self.CHANNELS = 2
self.RATE = 48000
self.CHUNK = self.RATE // chunk_rate
def printInfo(self):
def get_info(self):
dev_info = f"""
音频捕获进程：
- 捕获类型：{"音频输出" if self.audio_type == 0 else "音频输入"}
- 设备源：{self.source}
- 捕获进程PID：{self.process.pid if self.process else "None"}
- 捕获进程 PID：{self.process.pid if self.process else "None"}
音频样本块大小：{self.CHUNK}
样本格式：{self.FORMAT}
样本位宽：{self.SAMP_WIDTH}
采样格式：{self.FORMAT}
音频通道数：{self.CHANNELS}
音频采样率：{self.RATE}
样本通道数：{self.CHANNELS}
样本采样率：{self.RATE}
样本块大小：{self.CHUNK}
"""
print(dev_info)
def openStream(self):
def open_stream(self):
"""
启动音频捕获进程
"""
@@ -82,13 +86,23 @@ class AudioStream:
"""
读取音频数据
"""
if self.process:
if self.stop_signal:
self.close_stream()
return None
if self.process and self.process.stdout:
return self.process.stdout.read(self.CHUNK)
return None
def closeStream(self):
def close_stream_signal(self):
"""
线程安全的关闭系统音频输入流，不一定会立即关闭
"""
self.stop_signal = True
def close_stream(self):
"""
关闭系统音频捕获进程
"""
if self.process:
self.process.terminate()
self.stop_signal = False

57
engine/sysaudio/win.py
View File

@@ -1,14 +1,15 @@
"""获取 Windows 系统音频输入/输出流"""
import pyaudiowpatch as pyaudio
from textwrap import dedent
def getDefaultLoopbackDevice(mic: pyaudio.PyAudio, info = True)->dict:
def get_default_loopback_device(mic: pyaudio.PyAudio, info = True)->dict:
"""
获取默认的系统音频输出的回环设备
Args:
mic (pyaudio.PyAudio): pyaudio对象
info (bool, optional): 是否打印设备信息
mic: pyaudio对象
info: 是否打印设备信息
Returns:
dict: 系统音频输出的回环设备
@@ -51,38 +52,40 @@ class AudioStream:
self.audio_type = audio_type
self.mic = pyaudio.PyAudio()
if self.audio_type == 0:
self.device = getDefaultLoopbackDevice(self.mic, False)
self.device = get_default_loopback_device(self.mic, False)
else:
self.device = self.mic.get_default_input_device_info()
self.stop_signal = False
self.stream = None
self.SAMP_WIDTH = pyaudio.get_sample_size(pyaudio.paInt16)
self.INDEX = self.device["index"]
self.FORMAT = pyaudio.paInt16
self.SAMP_WIDTH = pyaudio.get_sample_size(self.FORMAT)
self.CHANNELS = int(self.device["maxInputChannels"])
self.RATE = int(self.device["defaultSampleRate"])
self.CHUNK = self.RATE // chunk_rate
self.INDEX = self.device["index"]
def printInfo(self):
def get_info(self):
dev_info = f"""
采样设备：
- 设备类型：{ "音频输出" if self.audio_type == 0 else "音频输入" }
- 序号：{self.device['index']}
- 名称：{self.device['name']}
- 设备序号：{self.device['index']}
- 设备名称：{self.device['name']}
- 最大输入通道数：{self.device['maxInputChannels']}
- 默认低输入延迟：{self.device['defaultLowInputLatency']}s
- 默认高输入延迟：{self.device['defaultHighInputLatency']}s
- 默认采样率：{self.device['defaultSampleRate']}Hz
- 是否回环设备：{self.device['isLoopbackDevice']}
音频样本块大小：{self.CHUNK}
设备序号：{self.INDEX}
样本格式：{self.FORMAT}
样本位宽：{self.SAMP_WIDTH}
采样格式：{self.FORMAT}
音频通道数：{self.CHANNELS}
音频采样率：{self.RATE}
样本通道数：{self.CHANNELS}
样本采样率：{self.RATE}
样本块大小：{self.CHUNK}
"""
print(dev_info)
return dedent(dev_info).strip()
def openStream(self):
def open_stream(self):
"""
打开并返回系统音频输出流
"""
@@ -96,18 +99,28 @@ class AudioStream:
)
return self.stream
def read_chunk(self):
def read_chunk(self) -> bytes | None:
"""
读取音频数据
"""
if self.stop_signal:
self.close_stream()
return None
if not self.stream: return None
return self.stream.read(self.CHUNK, exception_on_overflow=False)
def closeStream(self):
def close_stream_signal(self):
"""
关闭系统音频输出流
线程安全的关闭系统音频输入流，不一定会立即关闭
"""
if self.stream is None: return
self.stream.stop_stream()
self.stream.close()
self.stream = None
self.stop_signal = True
def close_stream(self):
"""
关闭系统音频输入流
"""
if self.stream is not None:
self.stream.stop_stream()
self.stream.close()
self.stream = None
self.stop_signal = False

2
engine/utils/__init__.py Normal file
View File

@@ -0,0 +1,2 @@
from .process import merge_chunk_channels, resample_chunk_mono, resample_mono_chunk
from .sysout import stdout, stdout_cmd, stdout_obj, stderr

22
engine/audioprcs/process.py → engine/utils/process.py
View File

@@ -1,16 +1,17 @@
import samplerate
import numpy as np
def mergeChunkChannels(chunk, channels):
def merge_chunk_channels(chunk: bytes, channels: int) -> bytes:
"""
将当前多通道音频数据块转换为单通道音频数据块
Args:
chunk: (bytes)多通道音频数据块
chunk: 多通道音频数据块
channels: 通道数
Returns:
(bytes)单通道音频数据块
单通道音频数据块
"""
# (length * channels,)
chunk_np = np.frombuffer(chunk, dtype=np.int16)
@@ -22,19 +23,19 @@ def mergeChunkChannels(chunk, channels):
return chunk_mono.tobytes()
def resampleRawChunk(chunk, channels, orig_sr, target_sr, mode="sinc_best"):
def resample_chunk_mono(chunk: bytes, channels: int, orig_sr: int, target_sr: int, mode="sinc_best") -> bytes:
"""
将当前多通道音频数据块转换成单通道音频数据块，然后进行重采样
Args:
chunk: (bytes)多通道音频数据块
chunk: 多通道音频数据块
channels: 通道数
orig_sr: 原始采样率
target_sr: 目标采样率
mode: 重采样模式，可选：'sinc_best' | 'sinc_medium' | 'sinc_fastest' | 'zero_order_hold' | 'linear'
Return:
(bytes)单通道音频数据块
单通道音频数据块
"""
# (length * channels,)
chunk_np = np.frombuffer(chunk, dtype=np.int16)
@@ -44,22 +45,23 @@ def resampleRawChunk(chunk, channels, orig_sr, target_sr, mode="sinc_best"):
chunk_mono_f = np.mean(chunk_np.astype(np.float32), axis=1)
chunk_mono = chunk_mono_f.astype(np.int16)
ratio = target_sr / orig_sr
chunk_mono_r = samplerate.resample(chunk_mono, ratio, converter_type=mode)
chunk_mono_r = samplerate.resample(chunk_mono, ratio, converter_type=mode)
chunk_mono_r = np.round(chunk_mono_r).astype(np.int16)
return chunk_mono_r.tobytes()
def resampleMonoChunk(chunk, orig_sr, target_sr, mode="sinc_best"):
def resample_mono_chunk(chunk: bytes, orig_sr: int, target_sr: int, mode="sinc_best") -> bytes:
"""
将当前单通道音频块进行重采样
Args:
chunk: (bytes)单通道音频数据块
chunk: 单通道音频数据块
orig_sr: 原始采样率
target_sr: 目标采样率
mode: 重采样模式，可选：'sinc_best' | 'sinc_medium' | 'sinc_fastest' | 'zero_order_hold' | 'linear'
Return:
(bytes)单通道音频数据块
单通道音频数据块
"""
chunk_np = np.frombuffer(chunk, dtype=np.int16)
ratio = target_sr / orig_sr

18
engine/utils/sysout.py Normal file
View File

@@ -0,0 +1,18 @@
import sys
import json
def stdout(text: str):
stdout_cmd("print", text)
def stdout_cmd(command: str, content = ""):
msg = { "command": command, "content": content }
sys.stdout.write(json.dumps(msg) + "\n")
sys.stdout.flush()
def stdout_obj(obj):
sys.stdout.write(json.dumps(obj) + "\n")
sys.stdout.flush()
def stderr(text: str):
sys.stderr.write(text + "\n")
sys.stderr.flush()