GetOffsetTimeV2 能用版本，校准的细节有待优化

Signed-off-by: allan716 <[email protected]>

internal/pkg/regex_things/regex_things.go

@@ -24,3 +24,6 @@ const regStringSRT = `(\d+)\n([\d:,]+)\s+-{2}\>\s+([\d:,]+)\n([\s\S]*?(\n{2}|$))
 
 var ReMatchDialogueSRT = regexp.MustCompile(regStringSRT)
 var ReMatchDialogueASS = regexp.MustCompile(regStringASS)
+
+// RegOneSeasonSubFolderNameMatch 每个视频文件夹下的缓存文件夹名称，一个季度的
+var RegOneSeasonSubFolderNameMatch = regexp.MustCompile(`(?m)^Sub_S\dE0`)

internal/pkg/sub_helper/sub_helper.go

@@ -7,16 +7,18 @@ import (
 	"github.com/allanpk716/ChineseSubFinder/internal/pkg/language"
 	"github.com/allanpk716/ChineseSubFinder/internal/pkg/log_helper"
 	"github.com/allanpk716/ChineseSubFinder/internal/pkg/my_util"
+	"github.com/allanpk716/ChineseSubFinder/internal/pkg/regex_things"
 	"github.com/allanpk716/ChineseSubFinder/internal/pkg/sub_parser_hub"
 	"github.com/allanpk716/ChineseSubFinder/internal/types/subparser"
 	"github.com/allanpk716/ChineseSubFinder/internal/types/supplier"
 	"github.com/go-rod/rod/lib/utils"
 	"io/ioutil"
+	"math"
 	"os"
 	"path/filepath"
-	"regexp"
 	"strconv"
 	"strings"
+	"time"
 )
 
 // OrganizeDlSubFiles 需要从汇总来是网站字幕中，解压对应的压缩包中的字幕出来
@@ -207,7 +209,7 @@ func SearchMatchedSubFileByDir(dir string) ([]string, error) {
 		fullPath := dir + pathSep + curFile.Name()
 		if curFile.IsDir() {
 			// 需要排除 Sub_S1E0、Sub_S2E0 这样的整季的字幕文件夹，这里仅仅是缓存，不会被加载的
-			matched := regOneSeasonSubFolderNameMatch.FindAllStringSubmatch(curFile.Name(), -1)
+			matched := regex_things.RegOneSeasonSubFolderNameMatch.FindAllStringSubmatch(curFile.Name(), -1)
 			if len(matched) > 0 {
 				continue
 			}
@@ -334,7 +336,7 @@ func DeleteOneSeasonSubCacheFolder(seriesDir string) error {
 	pathSep := string(os.PathSeparator)
 	for _, curFile := range files {
 		if curFile.IsDir() == true {
-			matched := regOneSeasonSubFolderNameMatch.FindAllStringSubmatch(curFile.Name(), -1)
+			matched := regex_things.RegOneSeasonSubFolderNameMatch.FindAllStringSubmatch(curFile.Name(), -1)
 			if matched == nil || len(matched) < 1 {
 				continue
 			}
@@ -376,6 +378,11 @@ func MergeMultiDialogue4EngSubtitle(inSubParser *subparser.FileInfo) {
 		目前看到的情况是前者的抽样率太低，需要使用后者的逻辑
 */
 func GetVADINfoFromSub(infoSrc *subparser.FileInfo, FrontAndEndPer float64, SubUnitMaxCount int) ([]SubUnit, error) {
+	return GetVADINfoFromSubNeedOffsetTime(infoSrc, FrontAndEndPer, SubUnitMaxCount, 0)
+}
+
+// GetVADINfoFromSubNeedOffsetTime 跟上面的函数功能一致，只不过这里可以加一个每一句话固定的偏移时间
+func GetVADINfoFromSubNeedOffsetTime(infoSrc *subparser.FileInfo, FrontAndEndPer float64, SubUnitMaxCount int, offsetTime float64) ([]SubUnit, error) {
 	if SubUnitMaxCount < 0 {
 		SubUnitMaxCount = 0
 	}
@@ -414,7 +421,12 @@ func GetVADINfoFromSub(infoSrc *subparser.FileInfo, FrontAndEndPer float64, SubU
 		}
 		// 低于 5句对白，则添加
 		if srcOneSubUnit.GetDialogueCount() < SubUnitMaxCount {
-			srcOneSubUnit.AddAndInsert(oneDialogueExTimeStart, oneDialogueExTimeEnd, i)
+			// 算上偏移
+			offsetTimeDuration := time.Duration(offsetTime * math.Pow10(9))
+			oneDialogueExTimeStart = oneDialogueExTimeStart.Add(offsetTimeDuration)
+			oneDialogueExTimeEnd = oneDialogueExTimeEnd.Add(offsetTimeDuration)
+			// 如果没有偏移就是 0
+			srcOneSubUnit.AddAndInsert(oneDialogueExTimeStart, oneDialogueExTimeEnd)
 		} else {
 			srcSubUnitList = append(srcSubUnitList, *srcOneSubUnit)
 			srcOneSubUnit = NewSubUnit()
@@ -427,7 +439,3 @@ func GetVADINfoFromSub(infoSrc *subparser.FileInfo, FrontAndEndPer float64, SubU
 
 	return srcSubUnitList, nil
 }
-
-var (
-	regOneSeasonSubFolderNameMatch = regexp.MustCompile(`(?m)^Sub_S\dE0`)
-)

internal/pkg/sub_helper/sub_unit.go

@@ -28,8 +28,8 @@ func NewSubUnit() *SubUnit {
 	}
 }
 
-// AddAndInsert 添加一句对白进来,并且填充中间的空白，间隔 10ms
-func (s *SubUnit) AddAndInsert(oneSubStartTime, oneSubEndTime time.Time, index int) {
+// AddAndInsert 添加一句对白进来,并且填充中间的空白，间隔 10ms。传入的时间是真实的时间
+func (s *SubUnit) AddAndInsert(oneSubStartTime, oneSubEndTime time.Time) {
 
 	/*
 		这里有个比较有意思的细节，字幕拆分到 dialogue 的时候，可能连续的多个 dialogue 是时间轴连续的
@@ -193,7 +193,7 @@ func (s SubUnit) GetOffsetTimeNumber() float64 {
 
 // GetFFMPEGCutRangeString 这里会生成导出 FFMPEG 的参数字段，起始时间和结束的时间长度
 // 以当前的 VAD 信息为基准，正负 expandTimeRange（秒为单位） 来生成截取的片段时间轴信息
-func (s SubUnit) GetFFMPEGCutRangeString(expandTimeRange float64) (string, string) {
+func (s SubUnit) GetFFMPEGCutRangeString(expandTimeRange float64) (string, string, time.Time, float64) {
 
 	var tmpStartTime time.Time
 	if s.GetStartTimeNumber(true)-expandTimeRange < 0 {
@@ -205,7 +205,9 @@ func (s SubUnit) GetFFMPEGCutRangeString(expandTimeRange float64) (string, strin
 	}
 
 	return fmt.Sprintf("%d:%d:%d.%d", tmpStartTime.Hour(), tmpStartTime.Minute(), tmpStartTime.Second(), tmpStartTime.Nanosecond()/1000/1000),
-		fmt.Sprintf("%f", s.GetTimelineRange()+expandTimeRange)
+		fmt.Sprintf("%f", s.GetTimelineRange()+expandTimeRange),
+		tmpStartTime,
+		s.GetTimelineRange() + expandTimeRange
 }
 
 // GetExpandRangeIndex 导出扩展的起始时间和结束的时间，整个多出的参数只适用于整体的字幕范围，局部不试用

internal/pkg/sub_timeline_fixer/fixer.go

@@ -380,16 +380,16 @@ func (s *SubTimelineFixer) GetOffsetTimeV2(infoBase, infoSrc *subparser.FileInfo
 	for _, srcSubUnit := range srcSubUnitList {
 
 		// 得到当前这个单元推算出来需要提取的字幕时间轴范围，这个是 Base Sub 使用的提取段
-		startTimeString, subLength := srcSubUnit.GetFFMPEGCutRangeString(ExpandTimeRange)
+		startTimeBaseString, subBaseLength, startTimeBaseTime, _ := srcSubUnit.GetFFMPEGCutRangeString(ExpandTimeRange)
 		// 导出当前的字幕文件适合与匹配的范围的临时字幕文件
-		nowTmpSubBaseFPath, errString, err := s.ffmpegHelper.ExportSubArgsByTimeRange(infoBase.FileFullPath, "base", startTimeString, subLength)
+		nowTmpSubBaseFPath, errString, err := s.ffmpegHelper.ExportSubArgsByTimeRange(infoBase.FileFullPath, "base", startTimeBaseString, subBaseLength)
 		if err != nil {
 			log_helper.GetLogger().Errorln("ExportSubArgsByTimeRange base", errString, err)
 			return false, 0, 0, err
 		}
 		// 导出当前的字幕文件适合与匹配的范围的临时字幕文件，这个是 Src Sub 使用的提取段
-		startTimeString, subLength = srcSubUnit.GetFFMPEGCutRangeString(0)
-		nowTmpSubSrcFPath, errString, err := s.ffmpegHelper.ExportSubArgsByTimeRange(infoSrc.FileFullPath, "src", startTimeString, subLength)
+		startTimeSrcString, subSrcLength, _, _ := srcSubUnit.GetFFMPEGCutRangeString(0)
+		nowTmpSubSrcFPath, errString, err := s.ffmpegHelper.ExportSubArgsByTimeRange(infoSrc.FileFullPath, "src", startTimeSrcString, subSrcLength)
 		if err != nil {
 			log_helper.GetLogger().Errorln("ExportSubArgsByTimeRange src", errString, err)
 			return false, 0, 0, err
@@ -403,6 +403,7 @@ func (s *SubTimelineFixer) GetOffsetTimeV2(infoBase, infoSrc *subparser.FileInfo
 			return false, 0, 0, errors.New("DetermineFileTypeFromFile == false")
 		}
 
+		// 这里比较特殊，因为读取的字幕文件是单独切割出来的，所以默认是有偏移的们需要使用不同的函数，把偏移算进去
 		nowTmpBaseSubUnitList, err := sub_helper.GetVADINfoFromSub(nowTmpSubBaseFileInfo, 0, 10000)
 		if err != nil {
 			return false, 0, 0, err
@@ -454,19 +455,36 @@ func (s *SubTimelineFixer) GetOffsetTimeV2(infoBase, infoSrc *subparser.FileInfo
 		correlationTM := treemap.NewWith(utils.Float64Comparator)
 		for i := 0; i < len(nowTmpBaseSubVADUnit.VADList); i++ {
 
-			correlation := CalculateCurveCorrelation(srcSubUnit.GetVADFloatSlice(), nowTmpBaseSubVADUnit.GetVADFloatSlice()[], len(srcSubUnit.VADList))
+			// 截取的长度是以当前 srcSubUnit 基准来判断的
+			// 类似滑动窗口的的功能实现
+			windowStartIndex := i
+			windowEndIndex := i + len(srcSubUnit.VADList)
+			if windowEndIndex >= len(nowTmpBaseSubVADUnit.VADList) {
+				break
+			}
+
+			correlation := CalculateCurveCorrelation(nowTmpBaseSubVADUnit.GetVADFloatSlice()[windowStartIndex:windowEndIndex], srcSubUnit.GetVADFloatSlice(), len(srcSubUnit.VADList))
 			correlationTM.Put(correlation, i)
-			println(fmt.Sprintf("%v - %v", i, correlation))
 		}
 		// 找到最大的数值和索引
-		_, tmpMaxIndex := correlationTM.Max() // tmpMaxCorrelation
+		tmpMaxCorrelation, tmpMaxIndex := correlationTM.Max() // tmpMaxCorrelation
+		if tmpMaxCorrelation == nil || tmpMaxIndex == nil {
+			continue
+		}
 		bok, nowBaseIndexTime := nowTmpBaseSubVADUnit.GetIndexTimeNumber(tmpMaxIndex.(int), true)
 		if bok == false {
 			continue
 		}
+		if tmpMaxCorrelation.(float64) <= MinCorrelation {
+			continue
+		}
+
 		nowSrcRealTime := srcSubUnit.GetStartTimeNumber(true)
 		// 时间差值
-		TimeDiffStart := nowBaseIndexTime - nowSrcRealTime
+		TimeDiffStart := nowBaseIndexTime + my_util.Time2SecendNumber(startTimeBaseTime) - nowSrcRealTime
+
+		println(fmt.Sprintf("%v <-> %v <-> %v", tmpMaxIndex, tmpMaxCorrelation, TimeDiffStart))
+
 		tmpStartDiffTime = append(tmpStartDiffTime, TimeDiffStart)
 		startDiffTimeList = append(startDiffTimeList, TimeDiffStart)
 	}
@@ -515,8 +533,8 @@ func (s *SubTimelineFixer) GetOffsetTimeV2(infoBase, infoSrc *subparser.FileInfo
 		newSd = oldSd
 	}
 
-	println(fmt.Sprintf("%v <-> %v <-> %v", oldMean, oldSd, per))
-	println(fmt.Sprintf("%v <-> %v <-> %v", newMean, newSd, per))
+	println(fmt.Sprintf("Old Mean: %v SD: %v Per: %v", oldMean, oldSd, per))
+	println(fmt.Sprintf("New Mean: %v SD: %v Per: %v", newMean, newSd, per))
 
 	return false, -1, -1, nil
 }
@@ -547,7 +565,7 @@ func (s *SubTimelineFixer) GetOffsetTimeV3(audioInfo vad.AudioInfo, infoSrc *sub
 	// 开始针对对白单元进行匹配
 	for _, subUnit := range subUnitList {
 
-		startTimeString, subLength := subUnit.GetFFMPEGCutRangeString(ExpandTimeRange)
+		startTimeString, subLength, _, _ := subUnit.GetFFMPEGCutRangeString(ExpandTimeRange)
 		// 导出当前的音频文件适合与匹配的范围的临时音频文件
 		outAudioFPath, _, errString, err := s.ffmpegHelper.ExportAudioAndSubArgsByTimeRange(audioInfo.FileFullPath, infoSrc.FileFullPath, startTimeString, subLength)
 		if err != nil {
@@ -634,7 +652,7 @@ func (s *SubTimelineFixer) GetOffsetTimeV3(audioInfo vad.AudioInfo, infoSrc *sub
 }
 
 const FixMask = "-fix"
-const FrontAndEndPer = 0.10 // 前百分之 15 和后百分之 15 都不进行识别
-const SubUnitMaxCount = 20  // 一个 Sub单元有五句对白
-const ExpandTimeRange = 40  // 从字幕的时间轴片段需要向前和向后多匹配一部分的音频，这里定义的就是这个 range 以分钟为单位， 正负 60 秒
-const MinCorelation = 0.4   // 最低的匹配度
+const FrontAndEndPer = 0.15 // 前百分之 15 和后百分之 15 都不进行识别
+const SubUnitMaxCount = 50  // 一个 Sub单元有五句对白
+const ExpandTimeRange = 50  // 从字幕的时间轴片段需要向前和向后多匹配一部分的音频，这里定义的就是这个 range 以分钟为单位， 正负 60 秒
+const MinCorrelation = 0.8  // 最低的匹配度