fixed some bugs

android/oral-trainer-sdk/src/main/java/cn/learningpad/oraltrainer/sdk/OralTrainerController.kt

@@ -10,7 +10,10 @@ import androidx.media3.common.C
 import androidx.media3.common.MediaItem
 import androidx.media3.common.PlaybackException
 import androidx.media3.common.Player
+import androidx.media3.common.TrackSelectionParameters.AudioOffloadPreferences
 import androidx.media3.exoplayer.ExoPlayer
+import androidx.media3.exoplayer.PlayerMessage
+import androidx.media3.exoplayer.SeekParameters
 import androidx.media3.exoplayer.source.DefaultMediaSourceFactory
 import java.util.concurrent.CopyOnWriteArraySet
 import kotlin.math.max
@@ -34,6 +37,11 @@ class OralTrainerController internal constructor(
     private var clippedSentence: SentenceBoundary? = null
     private var updatingPlaybackWindow = false
     private var unclippedDurationMs = -1L
+    private var sentenceEndMessage: PlayerMessage? = null
+    private var playFromStartAfterReady = false
+    private var clipStartNeedsWarmup = false
+    private var clipStartWarmupVolume: Float? = null
+    private val finishClipStartWarmupRunnable = Runnable { finishClipStartWarmup() }
 
     var config: PlayerConfig = initialConfig
         private set
@@ -48,6 +56,7 @@ class OralTrainerController internal constructor(
             if (released) {
                 return
             }
+            enforceStopAtSentenceEnd()
             notifySentenceIfChanged()
             notifySnapshot()
             mainHandler.postDelayed(this, IDLE_POLL_MS)
@@ -64,15 +73,38 @@ class OralTrainerController internal constructor(
             .also { exoPlayer ->
                 exoPlayer.repeatMode = Player.REPEAT_MODE_ALL
                 exoPlayer.playWhenReady = initialConfig.autoPlay
+                exoPlayer.setSeekParameters(SeekParameters.EXACT)
+                exoPlayer.setForegroundMode(true)
+                exoPlayer.trackSelectionParameters = exoPlayer.trackSelectionParameters
+                    .buildUpon()
+                    .setAudioOffloadPreferences(
+                        AudioOffloadPreferences.Builder()
+                            .setAudioOffloadMode(AudioOffloadPreferences.AUDIO_OFFLOAD_MODE_DISABLED)
+                            .build()
+                    )
+                    .build()
                 exoPlayer.addListener(object : Player.Listener {
                     override fun onPlaybackStateChanged(playbackState: Int) {
                         if (playbackState == Player.STATE_ENDED) {
-                            onClippedSentenceEnded()
+                            if (clipStartWarmupVolume != null) {
+                                finishClipStartWarmup()
+                            } else {
+                                onClippedSentenceEnded()
+                            }
                         }
-                        if (playbackState == Player.STATE_READY && unclippedDurationMs <= 0L) {
+                        if (playbackState == Player.STATE_READY) {
-                            val duration = player.duration
+                            if (playFromStartAfterReady) {
-                            if (duration != C.TIME_UNSET && clippedSentence == null) {
+                                playFromStartAfterReady = false
-                                unclippedDurationMs = duration
+                                playClippedSentenceFromStart()
+                            }
+                            if (stopAtSentenceEnd) {
+                                scheduleSentenceEndGuard()
+                            }
+                            if (unclippedDurationMs <= 0L) {
+                                val duration = player.duration
+                                if (duration != C.TIME_UNSET && clippedSentence == null) {
+                                    unclippedDurationMs = duration
+                                }
                             }
                         }
                         notifySnapshot()
@@ -148,6 +180,8 @@ class OralTrainerController internal constructor(
         lastSentenceIndex = null
         stopAtSentenceEnd = false
         stopAtSentenceEndIndex = null
+        cancelClipStartWarmup(restoreVolume = true)
+        cancelSentenceEndGuard()
         baseItemIndex = safeIndex
         clippedSentence = null
         unclippedDurationMs = -1L
@@ -193,12 +227,18 @@ class OralTrainerController internal constructor(
 
     fun play() {
         syncPlaybackWindow()
+        if (clippedSentence != null && !player.isPlaying && shouldWarmClipStart()) {
+            playClippedSentenceFromStart()
+            return
+        }
         player.play()
     }
 
     fun pause() {
         stopAtSentenceEnd = false
         stopAtSentenceEndIndex = null
+        cancelClipStartWarmup(restoreVolume = true)
+        cancelSentenceEndGuard()
         player.pause()
         applySentenceRepeatMode()
     }
@@ -214,12 +254,16 @@ class OralTrainerController internal constructor(
     fun stop() {
         stopAtSentenceEnd = false
         stopAtSentenceEndIndex = null
+        cancelClipStartWarmup(restoreVolume = true)
+        cancelSentenceEndGuard()
         player.stop()
     }
 
     fun seekTo(positionMs: Long) {
         stopAtSentenceEnd = false
         stopAtSentenceEndIndex = null
+        cancelClipStartWarmup(restoreVolume = true)
+        cancelSentenceEndGuard()
         val safePositionMs = max(0L, positionMs)
         lockSentenceAt(safePositionMs)
         syncPlaybackWindow(safePositionMs)
@@ -283,17 +327,25 @@ class OralTrainerController internal constructor(
 
     /** Seek to a known sentence and stop automatically when that sentence ends. */
     fun playSentenceAndStop(startMs: Long, sentenceIndex: Int) {
+        val item = currentTrainingItem()
+        val sentence = item?.sentences?.getOrNull(sentenceIndex)
+        cancelClipStartWarmup(restoreVolume = true)
         stopAtSentenceEnd = true
         stopAtSentenceEndIndex = sentenceIndex
         lastSentenceIndex = sentenceIndex
-        val item = currentTrainingItem()
+        player.playWhenReady = false
-        val sentence = item?.sentences?.getOrNull(sentenceIndex)
         if (item != null && sentence != null) {
-            installSentenceClip(item, sentence, 0L)
+            if (!isCurrentClip(sentence)) {
+                installSentenceClip(item, sentence, 0L)
+                player.playWhenReady = false
+            }
+            playClippedSentenceFromStart()
         } else {
+            cancelSentenceEndGuard()
             player.seekTo(max(0L, startMs))
+            player.playWhenReady = true
+            player.play()
         }
-        player.play()
         notifySentenceIfChanged(force = true)
         notifySnapshot()
     }
@@ -429,6 +481,8 @@ class OralTrainerController internal constructor(
             return
         }
         released = true
+        cancelClipStartWarmup(restoreVolume = false)
+        cancelSentenceEndGuard()
         mainHandler.removeCallbacks(ticker)
         listeners.clear()
         player.release()
@@ -498,17 +552,28 @@ class OralTrainerController internal constructor(
             return
         }
         rememberUnclippedDuration()
+        cancelSentenceEndGuard()
         val lengthMs = (sentence.endMs - sentence.startMs).coerceAtLeast(0L)
         val startPositionMs = relativePositionMs.coerceIn(0L, lengthMs)
         val playWhenReady = player.playWhenReady
+        val warmFromStart = startPositionMs <= CLIP_START_WARMUP_POSITION_MS
+        clipStartNeedsWarmup = warmFromStart
         updatingPlaybackWindow = true
         clippedSentence = sentence
         lastSentenceIndex = sentence.index
         player.setMediaItem(item.toMedia3Item(sentence.startMs, sentence.endMs), startPositionMs)
         player.prepare()
-        player.playWhenReady = playWhenReady
+        player.playWhenReady = false
         applySentenceRepeatMode()
         updatingPlaybackWindow = false
+        if (playWhenReady) {
+            if (warmFromStart) {
+                playClippedSentenceFromStart()
+            } else {
+                player.playWhenReady = true
+                player.play()
+            }
+        }
     }
 
     private fun restoreUnclipped(positionMs: Long) {
@@ -522,6 +587,7 @@ class OralTrainerController internal constructor(
         val playWhenReady = player.playWhenReady
         updatingPlaybackWindow = true
         clippedSentence = null
+        clipStartNeedsWarmup = false
         player.setMediaItems(
             mediaItems.map { it.toMedia3Item() },
             baseItemIndex.coerceIn(0, mediaItems.lastIndex),
@@ -545,16 +611,16 @@ class OralTrainerController internal constructor(
 
     private fun onClippedSentenceEnded() {
         val sentence = clippedSentence ?: return
+        if (stopAtSentenceEnd && stopAtSentenceEndIndex == sentence.index) {
+            handleStopAtSentenceEnd(sentence)
+            return
+        }
+        clipStartNeedsWarmup = false
         player.seekTo(0L)
         if (player.playbackState == Player.STATE_ENDED) {
             player.prepare()
         }
-        if (stopAtSentenceEnd && stopAtSentenceEndIndex == sentence.index) {
+        if (!config.continuousPlayback) {
-            stopAtSentenceEnd = false
-            stopAtSentenceEndIndex = null
-            player.pause()
-            applySentenceRepeatMode()
-        } else if (!config.continuousPlayback) {
             player.play()
         }
         lastSentenceIndex = sentence.index
@@ -562,6 +628,63 @@ class OralTrainerController internal constructor(
         notifySnapshot()
     }
 
+    private fun enforceStopAtSentenceEnd() {
+        if (!stopAtSentenceEnd) {
+            return
+        }
+        val index = stopAtSentenceEndIndex ?: return
+        val sentence = currentTrainingItem()?.sentences?.getOrNull(index) ?: return
+        if (playbackPositionMs() >= sentence.endMs) {
+            handleStopAtSentenceEnd(sentence)
+        }
+    }
+
+    private fun handleStopAtSentenceEnd(sentence: SentenceBoundary) {
+        if (!stopAtSentenceEnd || stopAtSentenceEndIndex != sentence.index) {
+            return
+        }
+        cancelSentenceEndGuard()
+        stopAtSentenceEnd = false
+        stopAtSentenceEndIndex = null
+        player.pause()
+        if (clippedSentence?.index == sentence.index) {
+            player.seekTo(0L)
+            if (player.playbackState == Player.STATE_ENDED) {
+                player.prepare()
+            }
+        } else {
+            seekInCurrentWindow(sentence.startMs)
+        }
+        applySentenceRepeatMode()
+        lastSentenceIndex = sentence.index
+        notifySentenceIfChanged(force = true)
+        notifySnapshot()
+    }
+
+    private fun scheduleSentenceEndGuard() {
+        val index = stopAtSentenceEndIndex ?: return
+        val sentence = currentTrainingItem()?.sentences?.getOrNull(index) ?: return
+        cancelSentenceEndGuard()
+        val endInWindowMs = if (clippedSentence?.index == sentence.index) {
+            (sentence.endMs - sentence.startMs - 1L).coerceAtLeast(0L)
+        } else {
+            sentence.endMs
+        }
+        sentenceEndMessage = player.createMessage { _, _ ->
+            if (stopAtSentenceEnd && stopAtSentenceEndIndex == sentence.index) {
+                handleStopAtSentenceEnd(sentence)
+            }
+        }.setLooper(mainHandler.looper)
+            .setPosition(endInWindowMs)
+            .setDeleteAfterDelivery(true)
+            .send()
+    }
+
+    private fun cancelSentenceEndGuard() {
+        sentenceEndMessage?.cancel()
+        sentenceEndMessage = null
+    }
+
     private fun switchBaseItem(targetIndex: Int) {
         if (targetIndex !in mediaItems.indices) {
             return
@@ -570,6 +693,8 @@ class OralTrainerController internal constructor(
         lastSentenceIndex = null
         stopAtSentenceEnd = false
         stopAtSentenceEndIndex = null
+        cancelClipStartWarmup(restoreVolume = true)
+        cancelSentenceEndGuard()
         clippedSentence = null
         unclippedDurationMs = -1L
         val item = mediaItems[targetIndex]
@@ -694,9 +819,93 @@ class OralTrainerController internal constructor(
         return CancellableAssessment {}
     }
 
+    private fun isCurrentClip(sentence: SentenceBoundary): Boolean {
+        val current = clippedSentence ?: return false
+        return current.index == sentence.index &&
+            current.startMs == sentence.startMs &&
+            current.endMs == sentence.endMs
+    }
+
+    private fun shouldWarmClipStart(): Boolean {
+        if (clippedSentence == null) {
+            return false
+        }
+        return clipStartNeedsWarmup || player.currentPosition <= CLIP_START_WARMUP_POSITION_MS
+    }
+
+    private fun playClippedSentenceFromStart() {
+        if (released) {
+            return
+        }
+        if (player.playbackState == Player.STATE_IDLE ||
+            player.playbackState == Player.STATE_ENDED ||
+            player.playbackState == Player.STATE_BUFFERING
+        ) {
+            playFromStartAfterReady = true
+            player.playWhenReady = false
+            if (player.playbackState == Player.STATE_IDLE || player.playbackState == Player.STATE_ENDED) {
+                player.prepare()
+            }
+            return
+        }
+        startClipStartWarmup()
+    }
+
+    private fun startClipStartWarmup() {
+        if (released) {
+            return
+        }
+        mainHandler.removeCallbacks(finishClipStartWarmupRunnable)
+        if (clipStartWarmupVolume == null) {
+            clipStartWarmupVolume = player.volume
+        }
+        player.volume = 0f
+        player.seekTo(0L)
+        if (player.playbackState == Player.STATE_ENDED) {
+            playFromStartAfterReady = true
+            player.prepare()
+            return
+        }
+        player.playWhenReady = true
+        player.play()
+        mainHandler.postDelayed(finishClipStartWarmupRunnable, CLIP_START_WARMUP_MS)
+    }
+
+    private fun finishClipStartWarmup() {
+        if (released) {
+            return
+        }
+        val restoreVolume = clipStartWarmupVolume ?: return
+        clipStartWarmupVolume = null
+        clipStartNeedsWarmup = false
+        mainHandler.removeCallbacks(finishClipStartWarmupRunnable)
+        player.seekTo(0L)
+        if (player.playbackState == Player.STATE_ENDED) {
+            player.prepare()
+        }
+        player.volume = restoreVolume
+        player.playWhenReady = true
+        player.play()
+        if (stopAtSentenceEnd) {
+            scheduleSentenceEndGuard()
+        }
+    }
+
+    private fun cancelClipStartWarmup(restoreVolume: Boolean) {
+        playFromStartAfterReady = false
+        mainHandler.removeCallbacks(finishClipStartWarmupRunnable)
+        val volume = clipStartWarmupVolume ?: return
+        clipStartWarmupVolume = null
+        if (restoreVolume) {
+            player.volume = volume
+        }
+    }
+
     private companion object {
         const val PREVIOUS_SENTENCE_TOLERANCE_MS = 250L
         const val NEXT_SENTENCE_TOLERANCE_MS = 150L
         const val IDLE_POLL_MS = 250L
+        const val CLIP_START_WARMUP_MS = 180L
+        const val CLIP_START_WARMUP_POSITION_MS = 120L
     }
 }

android/oral-trainer-sdk/src/main/java/cn/learningpad/oraltrainer/sdk/OralTrainerPlayerView.kt

@@ -16,6 +16,7 @@ class OralTrainerPlayerView @JvmOverloads constructor(
 ) : FrameLayout(context, attrs, defStyleAttr) {
     val playerView: PlayerView = PlayerView(context).apply {
         useController = false
+        setKeepContentOnPlayerReset(true)
         layoutParams = LayoutParams(LayoutParams.MATCH_PARENT, LayoutParams.MATCH_PARENT)
     }
 

android/sample-app/src/main/java/cn/learningpad/oraltrainer/sample/MainActivity.kt

@@ -449,8 +449,8 @@ class MainActivity : Activity() {
 
         override fun onReplaySentence() {
             testSession.stopDubbingIfNeeded()
-            testSession.refreshTestUi()
             controller.playCurrentSentenceAndStop()
+            testSession.refreshTestUi()
         }
 
         override fun onNextSentence() {

android/sample-app/src/main/java/cn/learningpad/oraltrainer/sample/SentenceExplanationOverlay.kt

@@ -57,7 +57,7 @@ internal class SentenceExplanationOverlay(
             setTextColor(SampleTheme.accentDeep)
             textSize = 14f
             setPadding(0, 8.dp, 0, 12.dp)
-            maxLines = 3
+            maxLines = 5
         }
         val message = TextView(activity).apply {
             text = "AI 正在讲解这句话…"

sentence_api/ai_tutor.py

@@ -54,8 +54,8 @@ class AITutor:
         language_note = f"句子语言：{language.strip()}。" if language.strip() else ""
         system_prompt = (
             "你是一位耐心的外语口语老师。请用简体中文讲解用户提供的影视/课程句子。"
-            "先给出自然的中文翻译，再解释关键词、常用搭配或语法要点，并简要说明这句话适合怎样跟读。"
+            "先给出自然的中文翻译，再解释关键词、常用搭配、语法要点，并简要说明这句话适合怎样跟读。"
-            "回答使用简明清晰的分段短文，总长度不超过 300 字，不要编造视频里没有的信息。"
+            "回答使用简明清晰的分段短文，总长度不超过 600 字，不要编造视频里没有的信息。"
         )
         user_prompt = (
             f"课程：{context}\n位置：{position}\n{language_note}句子：{sentence_text}"