From 60f05cebd8e7968621a0ce015cf72987ec7577c7 Mon Sep 17 00:00:00 2001
From: "Somhairle H. Marisol" <muqiu-han@outlook.com>
Date: Tue, 29 Sep 2026 10:50:12 +0800
Subject: feat(recorder): add offline voice clip pipeline
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

[变更性质]
- 本提交新增本机离线的语音片段录音能力，不涉及网络服务或语音识别。

[新增功能]
- 通过 Pulse 默认输入持续采集音频，以本地 Silero VAD 触发片段。
- 以私有目录和权限写入 24 kbps Ogg/Opus 录音，并提供 systemd 用户服务安装器。

[实现方案]
- 使用有限前置缓冲、静音封段和最长段滚动状态机，避免静音落盘及内存无限增长。
- 增加真实 Opus 编解码、服务 dry-run 和纯逻辑状态转换测试；记录模型归属和部署前置条件。

[影响范围]
- 新增 Python CLI、运行时模块、中文运维文档和自动测试。
- 未安装、启用或启动任何用户 systemd 服务；不删除或修改录音数据。
---
 tests/test_runtime.py | 82 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 82 insertions(+)
 create mode 100644 tests/test_runtime.py

(limited to 'tests/test_runtime.py')

diff --git a/tests/test_runtime.py b/tests/test_runtime.py
new file mode 100644
index 0000000..d0d45fb
--- /dev/null
+++ b/tests/test_runtime.py
@@ -0,0 +1,82 @@
+from datetime import datetime, timezone
+
+import numpy as np
+
+from mic_clipper.audio_input import PulseAudioInput
+from mic_clipper.runner import record_frames
+from mic_clipper.segmenter import Segmenter
+
+
+class FakeVad:
+    def __init__(self, probabilities: list[float]) -> None:
+        self.probabilities = iter(probabilities)
+
+    def probability(self, _samples: np.ndarray) -> float:
+        return next(self.probabilities)
+
+
+class FakeClip:
+    def __init__(self) -> None:
+        self.writes: list[np.ndarray] = []
+        self.closed = False
+
+    def write(self, samples: np.ndarray) -> None:
+        self.writes.append(samples)
+
+    def close(self) -> None:
+        self.closed = True
+
+    def abort(self) -> None:
+        self.closed = True
+
+
+class FakeWriter:
+    def __init__(self) -> None:
+        self.opened: list[FakeClip] = []
+
+    def open(self, _started_at: datetime) -> FakeClip:
+        clip = FakeClip()
+        self.opened.append(clip)
+        return clip
+
+
+def samples() -> np.ndarray:
+    return np.zeros(512, dtype=np.float32)
+
+
+def test_silence_never_opens_an_output_clip():
+    writer = FakeWriter()
+
+    record_frames(
+        [samples(), samples()],
+        vad=FakeVad([0.1, 0.1]),
+        writer=writer,
+        segmenter=Segmenter(pre_roll_seconds=0.5, silence_seconds=1.5),
+        now=lambda: datetime(2026, 9, 29, tzinfo=timezone.utc),
+    )
+
+    assert writer.opened == []
+
+
+def test_end_event_closes_the_active_encoder():
+    writer = FakeWriter()
+
+    record_frames(
+        [samples()] + [samples()] * 47,
+        vad=FakeVad([0.9] + [0.1] * 47),
+        writer=writer,
+        segmenter=Segmenter(pre_roll_seconds=0.5, silence_seconds=1.5),
+        now=lambda: datetime(2026, 9, 29, tzinfo=timezone.utc),
+    )
+
+    assert len(writer.opened) == 1
+    assert writer.opened[0].closed
+
+
+def test_pulse_input_uses_dynamic_default_device_at_16khz_mono():
+    command = PulseAudioInput.command()
+
+    assert command[0] == "ffmpeg"
+    assert command[command.index("-f") + 1] == "pulse"
+    assert command[command.index("-i") + 1] == "default"
+    assert command[-2:] == ["f32le", "pipe:1"]
-- 
cgit v1.2.3