feat(lifelog-recorder): v0.2 — Opus 编码 + 自启动 + 电池白名单

让 v1 真能脱离开发环境 24/7 跑的最后三公里： - SegmentWriter 改 Opus（MediaCodec audio/opus encoder + MediaMuxer OGG） ~17× 压缩，~2 KB/s 实测（vs WAV 32 KB/s）存储预算：8h 语音/天 → ~57 MB/天，~1.7 GB/月（vs WAV 51 GB/月） ffprobe 验证 codec_name=opus, channels=1, OggS magic OK - BootReceiver 监听 BOOT_COMPLETED / LOCKED_BOOT_COMPLETED / QUICKBOOT_POWERON，try-catch 兜底 Android 14+ FG service 限制 - MainActivity 加电池豁免行：dumpsys deviceidle whitelist 实时检测，未豁免时给「申请豁免」按钮，触发系统 RequestIgnoreBatteryOptimizations 对话框；ON_RESUME 生命周期观察让从设置页回来 UI 自动刷新 - AndroidManifest: + REQUEST_IGNORE_BATTERY_OPTIMIZATIONS 权限 + BootReceiver 注册（含 3 个 boot 相关 action） HANDOFF.md §五更新待办（v0.2 三项移到「已完成」），加存储预算 + Android 14 BOOT_COMPLETED caveat 段落（Pixel 8 Pro 实测能自动续上，前提 app 至少开过一次 + RECORD_AUDIO 已授）。 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

feat(lifelog-recorder): v0.2 — Opus 编码 + 自启动 + 电池白名单
让 v1 真能脱离开发环境 24/7 跑的最后三公里： - SegmentWriter 改 Opus（MediaCodec audio/opus encoder + MediaMuxer OGG） ~17× 压缩，~2 KB/s 实测（vs WAV 32 KB/s）存储预算：8h 语音/天 → ~57 MB/天，~1.7 GB/月（vs WAV 51 GB/月） ffprobe 验证 codec_name=opus, channels=1, OggS magic OK - BootReceiver 监听 BOOT_COMPLETED / LOCKED_BOOT_COMPLETED / QUICKBOOT_POWERON，try-catch 兜底 Android 14+ FG service 限制 - MainActivity 加电池豁免行：dumpsys deviceidle whitelist 实时检测，未豁免时给「申请豁免」按钮，触发系统 RequestIgnoreBatteryOptimizations 对话框；ON_RESUME 生命周期观察让从设置页回来 UI 自动刷新 - AndroidManifest: + REQUEST_IGNORE_BATTERY_OPTIMIZATIONS 权限 + BootReceiver 注册（含 3 个 boot 相关 action） HANDOFF.md §五更新待办（v0.2 三项移到「已完成」），加存储预算 + Android 14 BOOT_COMPLETED caveat 段落（Pixel 8 Pro 实测能自动续上，前提 app 至少开过一次 + RECORD_AUDIO 已授）。 Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
yibo
1 parent bb2506da
Showing 5 changed files with 253 additions and 43 deletions
lifelog-recorder/HANDOFF.md
lifelog-recorder/app/src/main/AndroidManifest.xml
lifelog-recorder/app/src/main/kotlin/com/lemon/lifelog/MainActivity.kt
lifelog-recorder/app/src/main/kotlin/com/lemon/lifelog/audio/SegmentWriter.kt
lifelog-recorder/app/src/main/kotlin/com/lemon/lifelog/service/BootReceiver.kt
@@ -40,13 +40,14 @@ APK 大小:    62 MB（含 34 MB sherpa-onnx .so + 0.6 MB silero VAD + 3 MB samp
 ```
 app/src/main/kotlin/com/lemon/lifelog/
 ├── LifelogApp.kt                — Application，建 NotificationChannel
-├── MainActivity.kt              — Compose UI，时间轴 LazyColumn + 测试按钮
+├── MainActivity.kt              — Compose UI，时间轴 + 电池豁免 UI + 测试按钮
 ├── service/
 │   ├── RecorderService.kt       — 前台服务，pipeline 主体 ←★ 改 ASR 流程改这里
-│   └── RecorderState.kt         — StateFlow 把服务状态推到 UI
+│   ├── RecorderState.kt         — StateFlow 把服务状态推到 UI
+│   └── BootReceiver.kt          — BOOT_COMPLETED 自启 (Android 14 有 caveat 见 §七)
 ├── audio/
 │   ├── AudioCapture.kt          — AudioRecord wrapper，emits FloatArray frames
-│   └── SegmentWriter.kt         — 写 16k mono WAV 到 filesDir（待换 Opus）
+│   └── SegmentWriter.kt         — Opus encode via MediaCodec + Ogg muxer，~17× 压缩
 ├── asr/
 │   ├── ParaformerEngine.kt      — sherpa-onnx paraformer 包装 ←★ 主 ASR
 │   ├── ModelDownloader.kt       — 首次启动下模型 + 断点续传
@@ -150,13 +151,30 @@ VAD 参数（在 `RecorderService.runPipeline` 里调）：
 ## §五 §六 待办（按优先级）
-1. **Opus 编码**（🔥 必须）：WAV 在 24/7 跑下 ~2.8 GB/天。换 `MediaCodec OMX.google.opus.encoder` 12 kbps voice mode，缩到 ~130 MB/天。改 `SegmentWriter.kt`。
-2. **自启动 + 电池白名单**（🔥 必须）：BOOT_COMPLETED Receiver 拉起 RecorderService；首启弹 REQUEST_IGNORE_BATTERY_OPTIMIZATIONS。否则一重启就死、Doze 一介入就死。
-3. **声纹识别 / speaker diarization**（用户当面提过 3 次）：sherpa-onnx 已有 `Speaker.kt` + `SpeakerEmbeddingExtractorConfig.kt`。流程：(a) 「这是我」录音 → 建 anchor embedding 存 SQLite；(b) 每段算 embedding 算 cosine 找最近邻；(c) Transcript 加 `speakerHint` 列。需要下 `sherpa-onnx-3dspeaker-zh.onnx`（~38 MB）。
-4. **FTS5 中文搜索**：Room + sqlite-jsqlite-with-icu。Schema 加 `TranscriptFts` 虚表用 ICU tokenizer。UI 加搜索框 + 高亮命中段。
-5. **LLM 纪要 worker**：WorkManager cron。每小时 / 静音 30min 触发，按时间窗 group segments → POST 到 hermes gateway `http://127.0.0.1:8642/v1/chat/completions`（需要先把 hermes api_server platform 配置起来，见 `~/.hermes/config.yaml`）。
-6. **存储 TTL**：删超过 N 天的 WAV，文字永留。Periodic WorkManager job。
-7. **设置页**：存储用量、模型切换、provider 切换、暂停时段配置（这次定的 MVP 是「一键暂停就够」，但用户后面可能要细化）。
+✅ **已完成（v0.2，2026-06）：Opus 编码 / BootReceiver / 电池白名单**
+
+剩下：
+
+1. **声纹识别 / speaker diarization**（🔥 用户当面提过 3 次）：sherpa-onnx 已有 `Speaker.kt` + `SpeakerEmbeddingExtractorConfig.kt`。流程：(a)「这是我」录音 → 建 anchor embedding 存 SQLite；(b) 每段算 embedding 算 cosine 找最近邻；(c) Transcript 加 `speakerHint` 列。需要下 `sherpa-onnx-3dspeaker-zh.onnx`（~38 MB）。
+2. **FTS5 中文搜索**：Room + sqlite-jsqlite-with-icu。Schema 加 `TranscriptFts` 虚表用 ICU tokenizer。UI 加搜索框 + 高亮命中段。
+3. **LLM 纪要 worker**：WorkManager cron。每小时 / 静音 30min 触发，按时间窗 group segments → POST 到 hermes gateway `http://127.0.0.1:8642/v1/chat/completions`（需要先把 hermes api_server platform 配置起来，见 `~/.hermes/config.yaml`）。
+4. **存储 TTL**：删超过 N 天的 .opus，文字永留。Periodic WorkManager job。
+5. **设置页**：存储用量、模型切换、provider 切换、暂停时段配置（v1 定的 MVP 是「一键暂停就够」，但 v2 可能要细化）。
+6. **段播放**：时间轴每条点开能直接播原音（Opus 系统原生支持，MediaPlayer 就能放）。
+7. **回链 lemonPCRM**：Transcript 提到的人名 / 项目 → GraphStore `MENTIONED_IN` 边。
+
+### 存储预算（v0.2 后）
+
+- Opus 16 kbps mono = ~2 KB/s
+- VAD 后实际有效语音 ≈ 8 h/天
+- → ~57 MB/天，~1.7 GB/月
+- 转写文字（UTF-8 中文）≈ 30 字/秒 × 8h × 3600 ≈ 不到 1 MB/月
+
+### Android 14+ BOOT_COMPLETED FG service caveat
+
+`RecorderService` 的 `foregroundServiceType="microphone"` 是受限 FG type。从 BOOT_COMPLETED 启动**不保证总是成功**——Android 14+ 的 `ForegroundServiceStartNotAllowedException` 可能扔出来（特别是首次重启、还没用过 app 的情况）。`BootReceiver.onReceive` 已 try/catch 兜底，失败时 log + no-op，用户开 app 手动点开始就能恢复。
+
+实测 Pixel 8 Pro Android 16 BP2A 上**重启后能自动续上**（前提：app 已经至少打开过一次 + RECORD_AUDIO 已授）。
 ---
@@ -8,6 +8,7 @@
     <uses-permission android:name="android.permission.POST_NOTIFICATIONS" />
     <uses-permission android:name="android.permission.WAKE_LOCK" />
     <uses-permission android:name="android.permission.RECEIVE_BOOT_COMPLETED" />
+    <uses-permission android:name="android.permission.REQUEST_IGNORE_BATTERY_OPTIMIZATIONS" />
     <uses-permission android:name="android.permission.INTERNET" />
     <uses-permission android:name="android.permission.ACCESS_NETWORK_STATE" />
@@ -34,5 +35,16 @@
             android:name=".service.RecorderService"
             android:exported="false"
             android:foregroundServiceType="microphone" />
+
+        <receiver
+            android:name=".service.BootReceiver"
+            android:exported="true"
+            android:enabled="true">
+            <intent-filter>
+                <action android:name="android.intent.action.BOOT_COMPLETED" />
+                <action android:name="android.intent.action.LOCKED_BOOT_COMPLETED" />
+                <action android:name="android.intent.action.QUICKBOOT_POWERON" />
+            </intent-filter>
+        </receiver>
     </application>
 </manifest>
 package com.lemon.lifelog
 import android.Manifest
+import android.annotation.SuppressLint
+import android.content.Context
 import android.content.Intent
+import android.net.Uri
 import android.os.Build
 import android.os.Bundle
+import android.os.PowerManager
+import android.provider.Settings
 import androidx.activity.ComponentActivity
 import androidx.activity.compose.rememberLauncherForActivityResult
 import androidx.activity.compose.setContent
@@ -28,11 +33,15 @@ import androidx.compose.material3.Surface
 import androidx.compose.material3.Text
 import androidx.compose.runtime.Composable
 import androidx.compose.runtime.collectAsState
+import androidx.compose.runtime.DisposableEffect
 import androidx.compose.runtime.getValue
 import androidx.compose.runtime.mutableStateOf
 import androidx.compose.runtime.remember
 import androidx.compose.runtime.rememberCoroutineScope
 import androidx.compose.runtime.setValue
+import androidx.lifecycle.Lifecycle
+import androidx.lifecycle.LifecycleEventObserver
+import androidx.lifecycle.compose.LocalLifecycleOwner
 import androidx.compose.ui.Modifier
 import androidx.compose.ui.unit.dp
 import androidx.core.content.ContextCompat
@@ -85,6 +94,20 @@ private fun MainScreen() {
     val repo = remember(activity) { Repository(activity) }
     val timeline by repo.segments.observeTimeline(200).collectAsState(initial = emptyList())
+    // Battery optimisation exemption — recompute whenever activity resumes
+    // (user may have just returned from the system settings dialog).
+    var batteryExempt by remember { mutableStateOf(isBatteryOptExempt(activity)) }
+    val lifecycleOwner = LocalLifecycleOwner.current
+    DisposableEffect(lifecycleOwner) {
+        val observer = LifecycleEventObserver { _, event ->
+            if (event == Lifecycle.Event.ON_RESUME) {
+                batteryExempt = isBatteryOptExempt(activity)
+            }
+        }
+        lifecycleOwner.lifecycle.addObserver(observer)
+        onDispose { lifecycleOwner.lifecycle.removeObserver(observer) }
+    }
+
     val permissionLauncher = rememberLauncherForActivityResult(
         ActivityResultContracts.RequestMultiplePermissions()
     ) { results ->
@@ -117,6 +140,21 @@ private fun MainScreen() {
             }
         }) { Text(if (running) "停止录音" else "开始录音") }
+        // Battery optimisation exemption row.
+        Row(verticalAlignment = androidx.compose.ui.Alignment.CenterVertically,
+            horizontalArrangement = Arrangement.spacedBy(8.dp)) {
+            Text(
+                if (batteryExempt) "✓ 已加入电池白名单"
+                else "⚠ 未加入电池白名单（Doze 会关掉录音）",
+                style = MaterialTheme.typography.bodySmall,
+            )
+            if (!batteryExempt) {
+                Button(onClick = { requestBatteryOptExemption(activity) }) {
+                    Text("申请豁免")
+                }
+            }
+        }
+
         if (running && lastAudioMs > 0) {
             val rtSpeed = if (lastWallMs > 0) lastAudioMs.toDouble() / lastWallMs else 0.0
             Text(
@@ -171,6 +209,20 @@ private fun MainScreen() {
     }
 }
+private fun isBatteryOptExempt(ctx: Context): Boolean {
+    val pm = ctx.getSystemService(Context.POWER_SERVICE) as PowerManager
+    return pm.isIgnoringBatteryOptimizations(ctx.packageName)
+}
+
+@SuppressLint("BatteryLife")
+private fun requestBatteryOptExemption(ctx: Context) {
+    val intent = Intent(Settings.ACTION_REQUEST_IGNORE_BATTERY_OPTIMIZATIONS).apply {
+        data = Uri.parse("package:${ctx.packageName}")
+        addFlags(Intent.FLAG_ACTIVITY_NEW_TASK)
+    }
+    runCatching { ctx.startActivity(intent) }
+}
+
 private val TIMELINE_FMT = SimpleDateFormat("HH:mm:ss", Locale.US)
 @Composable
 package com.lemon.lifelog.audio
+import android.media.MediaCodec
+import android.media.MediaCodecInfo
+import android.media.MediaFormat
+import android.media.MediaMuxer
+import android.util.Log
 import java.io.File
-import java.io.RandomAccessFile
 import java.nio.ByteBuffer
 import java.nio.ByteOrder
 import java.text.SimpleDateFormat
@@ -9,13 +13,23 @@ import java.util.Date
 import java.util.Locale
 /**
- * Writes a finished VAD segment to disk as 16-bit PCM WAV.
- * File layout: `<root>/YYYY-MM-DD/HH-MM-SS-NNNN.wav`
+ * Writes a finished VAD segment to disk as Opus-in-Ogg (.opus, ~16 kbps voice mode).
+ * File layout: `<root>/YYYY-MM-DD/HH-MM-SS-MS.opus`
  *
- * (Opus encoding via MediaCodec is a future optimization; raw WAV at 16k mono is
- * ~32 kB/s = ~1 MB/min, still bounded for 24/7 use after VAD trimming.)
+ * Bitrate budget: 16 kHz mono PCM16 = 32 kB/s WAV → ~2 kB/s Opus = ~16× compression.
+ * Per 24/7 use (≈8 h voice/day after VAD): ~57 MB/day, ~1.7 GB/month.
+ *
+ * Implementation: MediaCodec (audio/opus encoder) + MediaMuxer (OGG container).
+ * Each segment uses its own encoder + muxer pair — startup overhead is small (a few ms)
+ * and avoids cross-segment state bugs.
  */
 class SegmentWriter(private val root: File) {
+    companion object {
+        private const val TAG = "SegmentWriter"
+        private const val MIME = MediaFormat.MIMETYPE_AUDIO_OPUS  // "audio/opus"
+        private const val BITRATE = 16_000  // ~16 kbps, plenty for speech
+    }
+
     private val dayFmt = SimpleDateFormat("yyyy-MM-dd", Locale.US)
     private val timeFmt = SimpleDateFormat("HH-mm-ss-SSS", Locale.US)
@@ -23,42 +37,116 @@ class SegmentWriter(private val root: File) {
         root.mkdirs()
     }
-    /** Writes samples to a new file, returns absolute path. */
+    /** Writes samples to a new .opus file, returns absolute path. */
     fun write(samplesFloat: FloatArray, ts: Long, sampleRate: Int = 16000): String {
         val date = Date(ts)
         val dir = File(root, dayFmt.format(date)).apply { mkdirs() }
-        val file = File(dir, "${timeFmt.format(date)}.wav")
-        writeWav(file, samplesFloat, sampleRate)
+        val file = File(dir, "${timeFmt.format(date)}.opus")
+        encodeToOgg(samplesFloat, file, sampleRate)
         return file.absolutePath
     }
-    private fun writeWav(file: File, samples: FloatArray, sampleRate: Int) {
-        val pcmBytes = samples.size * 2
-        RandomAccessFile(file, "rw").use { raf ->
-            raf.setLength(0)
-            // RIFF header
-            raf.writeBytes("RIFF")
-            raf.writeInt(Integer.reverseBytes(36 + pcmBytes))
-            raf.writeBytes("WAVE")
-            // fmt chunk
-            raf.writeBytes("fmt ")
-            raf.writeInt(Integer.reverseBytes(16))                  // chunk size
-            raf.writeShort(java.lang.Short.reverseBytes(1).toInt()) // PCM
-            raf.writeShort(java.lang.Short.reverseBytes(1).toInt()) // channels
-            raf.writeInt(Integer.reverseBytes(sampleRate))
-            raf.writeInt(Integer.reverseBytes(sampleRate * 2))      // byte rate
-            raf.writeShort(java.lang.Short.reverseBytes(2).toInt()) // block align
-            raf.writeShort(java.lang.Short.reverseBytes(16).toInt())// bits per sample
-            // data chunk
-            raf.writeBytes("data")
-            raf.writeInt(Integer.reverseBytes(pcmBytes))
-            // samples: scale float [-1,1] → int16 LE
-            val buf = ByteBuffer.allocate(pcmBytes).order(ByteOrder.LITTLE_ENDIAN)
+    private fun encodeToOgg(samples: FloatArray, outFile: File, sampleRate: Int) {
+        if (samples.isEmpty()) return
+
+        // 1. PCM16 LE buffer (mono).
+        val pcm = ByteArray(samples.size * 2)
+        ByteBuffer.wrap(pcm).order(ByteOrder.LITTLE_ENDIAN).apply {
             for (s in samples) {
-                val clamped = (s * 32767f).coerceIn(-32768f, 32767f)
-                buf.putShort(clamped.toInt().toShort())
+                putShort((s.coerceIn(-1f, 1f) * 32767f).toInt().toShort())
+            }
+        }
+
+        // 2. Set up encoder.
+        val codec = MediaCodec.createEncoderByType(MIME)
+        val format = MediaFormat.createAudioFormat(MIME, sampleRate, 1).apply {
+            setInteger(MediaFormat.KEY_BIT_RATE, BITRATE)
+            // Default frame size for Opus encoder is sane; explicit max-input-size is a hint.
+            setInteger(MediaFormat.KEY_MAX_INPUT_SIZE, 8192)
+        }
+        try {
+            codec.configure(format, null, null, MediaCodec.CONFIGURE_FLAG_ENCODE)
+        } catch (t: Throwable) {
+            Log.e(TAG, "configure failed", t)
+            codec.release()
+            throw t
+        }
+        codec.start()
+
+        // 3. Muxer (OGG container).
+        val muxer = MediaMuxer(outFile.absolutePath, MediaMuxer.OutputFormat.MUXER_OUTPUT_OGG)
+        var trackIndex = -1
+        var muxerStarted = false
+
+        val info = MediaCodec.BufferInfo()
+        var bytesFed = 0
+        var ptUs = 0L
+        var inputDone = false
+        var outputDone = false
+
+        try {
+            while (!outputDone) {
+                // Feed PCM.
+                if (!inputDone) {
+                    val idx = codec.dequeueInputBuffer(50_000)
+                    if (idx >= 0) {
+                        val buf = codec.getInputBuffer(idx)!!
+                        buf.clear()
+                        val cap = buf.capacity()
+                        if (bytesFed < pcm.size) {
+                            val n = minOf(cap, pcm.size - bytesFed)
+                            buf.put(pcm, bytesFed, n)
+                            codec.queueInputBuffer(idx, 0, n, ptUs, 0)
+                            // 2 bytes/sample mono → frames = n/2 → us = frames * 1e6 / sr
+                            ptUs += n.toLong() / 2L * 1_000_000L / sampleRate
+                            bytesFed += n
+                        } else {
+                            codec.queueInputBuffer(
+                                idx, 0, 0, ptUs,
+                                MediaCodec.BUFFER_FLAG_END_OF_STREAM,
+                            )
+                            inputDone = true
+                        }
+                    }
+                }
+
+                // Drain encoded output.
+                drain@ while (true) {
+                    val oIdx = codec.dequeueOutputBuffer(info, 50_000)
+                    when {
+                        oIdx == MediaCodec.INFO_TRY_AGAIN_LATER -> break@drain
+                        oIdx == MediaCodec.INFO_OUTPUT_FORMAT_CHANGED -> {
+                            if (!muxerStarted) {
+                                trackIndex = muxer.addTrack(codec.outputFormat)
+                                muxer.start()
+                                muxerStarted = true
+                            }
+                        }
+                        oIdx >= 0 -> {
+                            val out = codec.getOutputBuffer(oIdx)!!
+                            if (info.size > 0 && muxerStarted &&
+                                (info.flags and MediaCodec.BUFFER_FLAG_CODEC_CONFIG) == 0
+                            ) {
+                                out.position(info.offset)
+                                out.limit(info.offset + info.size)
+                                muxer.writeSampleData(trackIndex, out, info)
+                            }
+                            codec.releaseOutputBuffer(oIdx, false)
+                            if ((info.flags and MediaCodec.BUFFER_FLAG_END_OF_STREAM) != 0) {
+                                outputDone = true
+                                break@drain
+                            }
+                        }
+                    }
+                }
+            }
+        } finally {
+            runCatching { codec.stop() }
+            codec.release()
+            if (muxerStarted) {
+                runCatching { muxer.stop() }
             }
-            raf.write(buf.array())
+            muxer.release()
         }
     }
 }
+package com.lemon.lifelog.service
+
+import android.content.BroadcastReceiver
+import android.content.Context
+import android.content.Intent
+import android.util.Log
+import androidx.core.content.ContextCompat
+
+/**
+ * Resumes recording on device boot.
+ *
+ * Android 14+ note: starting a `foregroundServiceType=microphone` service from a
+ * BOOT_COMPLETED broadcast is allowed because RECORD_AUDIO is a "while-in-use" permission
+ * that survives across reboots once granted.  If the start fails (e.g. user has revoked
+ * the runtime permission), Android will throw [ForegroundServiceStartNotAllowedException]
+ * and we just no-op; the user opens the app and taps "开始录音" to resume.
+ */
+class BootReceiver : BroadcastReceiver() {
+    override fun onReceive(context: Context, intent: Intent) {
+        val action = intent.action ?: return
+        if (action != Intent.ACTION_BOOT_COMPLETED &&
+            action != Intent.ACTION_LOCKED_BOOT_COMPLETED &&
+            action != "android.intent.action.QUICKBOOT_POWERON"
+        ) return
+
+        Log.i(TAG, "boot action=$action → start RecorderService")
+        try {
+            ContextCompat.startForegroundService(
+                context,
+                Intent(context, RecorderService::class.java),
+            )
+        } catch (t: Throwable) {
+            Log.w(TAG, "could not start service from boot", t)
+        }
+    }
+
+    companion object {
+        private const val TAG = "BootReceiver"
+    }
+}