文章目录
- 0、参考资料
- 1、WAV 格式了解
- 1.1 WAV 文件头
- 1.2 RIFF Chunk 区块
- 1.3 Format Chunk 区块
- 1.4 Data Chunk 区块
- 2、音频剪裁 -> 解码 -> 编码
- 2.1 mp3
- 2.1.1 裁剪
- 2.1.2 解码
- 2.1.3 编码
- 2.2 pcm 裁剪
0、参考资料
【音频处理】WAV 文件格式分析 ( 逐个字节解析文件头 | 相关字段的计算公式 )
android音频学习笔记之wav头文件
1、WAV 格式了解
WAV格式，多媒体中使用的声波文件格式之一，它是以RIFF格式为标准的。RIFF是英文 Resource Interchange File Format 的缩写，每个WAV文件的头四个字节便是“RIFF”。WAV文件分为两部分：- 第一部分：文件头，记录重要的参数信息，对于音频而言，包括采样率、通道数、位宽等
- 第二部分：数据块，也就是一帧一帧的二进制数据，对于音频而言，就是原始的
PCM数据
1.1 WAV 文件头
WAV文件头，主要分为三部分：RIFF Chunk 区块、Format Chunk 区块、Data Chunk 区块
1.2 RIFF Chunk 区块
RIFF数据块长度为 12 字节，共有三种码段。ChunkID：填入“RIFF”为标识，表示这是一个RIFF格式文件ChunkSize：除了RIFF及自己以外，整个文件的长度，该值计算方式为 PCM 音频样本总字节数 + 44 − 8Format：填入WAVE来标识这是一个wav文件
// RIFF 资源交换文件标志
header[0] = 'R'.code.toByte()
header[1] = 'I'.code.toByte()
header[2] = 'F'.code.toByte()
header[3] = 'F'.code.toByte()
// 数据大小
header[4] = (totalDataLen and 0xffL).toByte()
header[5] = (totalDataLen shr 8 and 0xffL).toByte()
header[6] = (totalDataLen shr 16 and 0xffL).toByte()
header[7] = (totalDataLen shr 24 and 0xffL).toByte()
// WAVE 标识 wav 文件
header[8] = 'W'.code.toByte()
header[9] = 'A'.code.toByte()
header[10] = 'V'.code.toByte()
header[11] = 'E'.code.toByte()
1.3 Format Chunk 区块
-
Subchunk1ID：填入“fmt”为标识，波形格式标志 -
Subchunk1Size：Subchunk1减去Subchunk1ID和Subchunk1Size之后剩下的长度，一般为 16 -
AudioFormat：编码格式，即压缩格式，1 表示PCM格式，无压缩wav文件几乎支持所有ACM规范的编码格式，其信息存储在文件头 21、22 两个字节中，有如下编码格式：
-
NumChannels：声道数，1 表示单声道，2 表示双声道 -
SampleRate：采样率 -
ByteRate：波形数据传输速率, 即每秒中的字节数，计算公式为：
-
BlockAlign：标识块对齐的内容（数据块的调整数），播放软件一次处理多少个该值大小的字节数据，以便将其用于缓冲区的调整，也标识一帧的字节数。每个采样所需的字节数，计算公式为：
-
BitsPerSample：采样位宽，即每个采样点的bit数，其中 8 表示 8bit，16 表示 16bit，32 表示 32bit。
// 'fmt '：波形格式标志,最后一位空格
header[12] = 'f'.code.toByte()
header[13] = 'm'.code.toByte()
header[14] = 't'.code.toByte()
header[15] = ' '.code.toByte() // 过渡字节
// 数据大小
header[16] = 16 // 4 bytes: size of 'fmt ' chunk
header[17] = 0
header[18] = 0
header[19] = 0
// 编码方式为 PCM 编码格式
header[20] = 1 // format = 1
header[21] = 0
// 通道数
header[22] = channels.toByte()
header[23] = 0
// 采样率，每个通道的播放速度
header[24] = (sampleRate and 0xff).toByte()
header[25] = (sampleRate shr 8 and 0xff).toByte()
header[26] = (sampleRate shr 16 and 0xff).toByte()
header[27] = (sampleRate shr 24 and 0xff).toByte()
// 音频数据传送速率，采样率 * 通道数 * 采样深度 / 8
header[28] = (byteRate and 0xffL).toByte()
header[29] = (byteRate shr 8 and 0xffL).toByte()
header[30] = (byteRate shr 16 and 0xffL).toByte()
header[31] = (byteRate shr 24 and 0xffL).toByte()
// 确定系统一次要处理多少个这样字节的数据，确定缓冲区，通道数 * 采样位数
header[32] = (channels * 16 / 8).toByte()
header[33] = 0
// 每个样本的数据位数
header[34] = 16
header[35] = 0
1.4 Data Chunk 区块
Subchunk2ID：填入“data”为标识Subchunk2Size：PCM音频数据的总长度，单位字节，即文件总字节数 - 44
// Data chunk
header[36] = 'd'.code.toByte() // data
header[37] = 'a'.code.toByte()
header[38] = 't'.code.toByte()
header[39] = 'a'.code.toByte()
header[40] = (totalAudioLen and 0xffL).toByte()
header[41] = (totalAudioLen shr 8 and 0xffL).toByte()
header[42] = (totalAudioLen shr 16 and 0xffL).toByte()
header[43] = (totalAudioLen shr 24 and 0xffL).toByte()
2、音频剪裁 -> 解码 -> 编码
2.1 mp3
2.1.1 裁剪
/*** mp3 音频文件截取工具*/
object Mp3CutUtils {private const val TAG = "Mp3CutUtils"/* 适当的调整 SAMPLE_SIZE 可以更加精确的裁剪音乐 */private const val SAMPLE_SIZE = 1024 * 200/*微秒*/private const val MICRO_SECOND = 1000_000/*** 裁剪 MP3 格式的音频* @param sourceFile 源文件地址* @param targetFile 目标文件地址* @param startSecond 截取开始时间（秒）* @param endSecond 截取结束时间（秒）*/fun cutMp3Audio(sourceFile: String,targetFile: String,startSecond: Float,endSecond: Float): Boolean {// 转换为微秒val startMicrosecond = startSecond * MICRO_SECONDval endMicrosecond = endSecond * MICRO_SECONDMLog.d(TAG,"cutMp3Audio: sourceFile = $sourceFile, targetFile = $targetFile, startMicrosecond = $startMicrosecond, endMicrosecond = $endMicrosecond")var extractor: MediaExtractor? = nullvar outputStream: BufferedOutputStream? = nulltry {extractor = MediaExtractor()extractor.setDataSource(sourceFile)val track = getAudioTrack(extractor)if (track < 0) {return false}// 选择音频轨道extractor.selectTrack(track)outputStream = BufferedOutputStream(FileOutputStream(targetFile), SAMPLE_SIZE)MLog.d(TAG, "cutMp3Audio: startMicrosecond = $startMicrosecond, endMicrosecond = $endMicrosecond")//跳至开始裁剪位置extractor.seekTo(startMicrosecond.toLong(), MediaExtractor.SEEK_TO_PREVIOUS_SYNC)while (true) {val buffer: ByteBuffer = ByteBuffer.allocate(SAMPLE_SIZE)val sampleSize = extractor.readSampleData(buffer, 0)val timeStamp = extractor.sampleTimeMLog.d(TAG, "cutMp3Audio: sampleSize = $sampleSize, timeStamp = $timeStamp")// >= 1000000 是要裁剪停止和指定的裁剪结尾不小于 1 秒，否则裁剪 5 秒可能产生需要 4 秒音频// 裁剪到只有 4.6 秒，大多数音乐播放器是向下取整，这样对于播放器变成了 4 秒，// 所以要裁剪比 4 秒多一秒的边界if (timeStamp > endMicrosecond && timeStamp - endMicrosecond >= MICRO_SECOND) {break}if (sampleSize <= 0) {break}val buf = ByteArray(sampleSize)buffer.get(buf, 0, sampleSize)// 写入文件outputStream.write(buf)// 音轨数据往前读extractor.advance()}} catch (e: Exception) {e.printStackTrace()return false} finally {extractor?.release()if (outputStream != null) {try {outputStream.close()} catch (e: Exception) {e.printStackTrace()}}}return true}/*** 获取音频数据轨道* @param audioExtractor*/fun getAudioTrack(audioExtractor: MediaExtractor): Int {for (i in 0 until audioExtractor.trackCount) {val format = audioExtractor.getTrackFormat(i)val mime = format.getString(MediaFormat.KEY_MIME)if (mime?.startsWith(AudioConstant.AUDIO_PREFIX) == true) {return i}}return -1}}
2.1.2 解码
/*** MP3 -> PCM 解码*/
object Mp3DecodeUtil {private const val TAG = "Mp3DecodeUtil"/*** 解码 MP3 文件* @param mp3Path mp3 文件路径*/fun decodeMP3(mp3Path: String): Boolean {MLog.d(TAG, "decodeMP3: mp3Path = $mp3Path")val audioExtractor = MediaExtractor()var audioDecoder: MediaCodec? = nulltry {audioExtractor.setDataSource(mp3Path)val track = getAudioTrack(audioExtractor)if (track != -1) {val format = audioExtractor.getTrackFormat(track)val mime = format.getString(MediaFormat.KEY_MIME)audioExtractor.selectTrack(track)audioDecoder = MediaCodec.createDecoderByType(mime!!)audioDecoder.configure(format, null, null, 0)audioDecoder.start()}} catch (e: IOException) {e.printStackTrace()return false}coreDecode(mp3Path, audioExtractor, audioDecoder)MLog.d(TAG, "decoderMP3: stop")audioDecoder?.stop()audioDecoder?.release()audioExtractor?.release()return true}private fun coreDecode(mp3:String, audioExtractor: MediaExtractor, audioDecoder: MediaCodec?) {audioDecoder ?: returnval info = MediaCodec.BufferInfo()// 得到输出 PCM 文件的路径 val pcmFilePath =val pcmFilePath = "${mp3.substring(0, mp3.lastIndexOf("."))}${AudioConstant.SUFFIX_PCM}"val fos = FileOutputStream(File(pcmFilePath))val bufferedOutputStream = BufferedOutputStream(fos)while (true) {extractorInputBuffer(audioExtractor, audioDecoder!!)val outIndex = audioDecoder.dequeueOutputBuffer(info, 50000)MLog.d(TAG, "decoderMP3: presentationTimeUs = ${info.presentationTimeUs}, outIndex = $outIndex")if (outIndex >= 0) {MLog.d(TAG, "decoderMP3: outIndex >= 0")val data = audioDecoder.getOutputBuffer(outIndex)if (info.flags and MediaCodec.BUFFER_FLAG_CODEC_CONFIG != 0) {info.size = 0}if (info.size != 0) {data!!.position(info.offset)data.limit(info.offset + info.size)// 将数据写入 bufferedOutputStream 中bufferedOutputStream.write(ByteArray(data.remaining()) { data.get() })data.clear()}audioDecoder.releaseOutputBuffer(outIndex, false)}if (info.flags and MediaCodec.BUFFER_FLAG_END_OF_STREAM != 0) {MLog.d(TAG, "decoderMP3: MediaCodec.BUFFER_FLAG_END_OF_STREAM) != 0")convertPcm2Wav(pcmFilePath, audioExtractor)break}}}private fun extractorInputBuffer(mediaExtractor: MediaExtractor, mediaCodec: MediaCodec) {val inputIndex = mediaCodec.dequeueInputBuffer(50000)if (inputIndex >= 0) {val inputBuffer = mediaCodec.getInputBuffer(inputIndex)val sampleTime = mediaExtractor.sampleTimeval sampleSize = mediaExtractor.readSampleData(inputBuffer!!, 0)if (mediaExtractor.advance()) {mediaCodec.queueInputBuffer(inputIndex, 0, sampleSize, sampleTime, 0)} else {if (sampleSize > 0) {mediaCodec.queueInputBuffer(inputIndex,0,sampleSize,sampleTime,MediaCodec.BUFFER_FLAG_END_OF_STREAM)} else {mediaCodec.queueInputBuffer(inputIndex,0,0,0,MediaCodec.BUFFER_FLAG_END_OF_STREAM)}}}}private fun convertPcm2Wav(pcmFilePath: String, audioExtractor: MediaExtractor) {val wavFilePath = "${pcmFilePath.substring(0, pcmFilePath.lastIndexOf("."))}${AudioConstant.SUFFIX_CUT_WAV}"MLog.d(TAG, "convertPcm2Wav: pcmFilePath = $pcmFilePath, wavFilePath = $wavFilePath")val track = getAudioTrack(audioExtractor)if (track == -1) returnval format = audioExtractor.getTrackFormat(track)val duration = format.getLong(MediaFormat.KEY_DURATION)val channelCount = format.getInteger(MediaFormat.KEY_CHANNEL_COUNT)val sampleRate = format.getInteger(MediaFormat.KEY_SAMPLE_RATE)val byteNumber = (if (format.containsKey(MediaFormat.KEY_WIDTH)) format.getInteger("bit-width") else 16) / 8MLog.d(TAG, "convertPcm2Wav: duration = $duration, sampleRate = $sampleRate, channelCount = $channelCount, byteNumber = $byteNumber")PcmEncodeUtil.convertPcm2Wav(pcmFilePath, wavFilePath, sampleRate, channelCount, byteNumber)}
}
2.1.3 编码
/*** PCM -> WAV 编码*/
object PcmEncodeUtil {private const val TAG = "PcmEncodeUtil"/*** PCM 文件转 WAV 文件* @param inPcmFilePath 输入 PCM 文件路径* @param outWavFilePath 输出 WAV 文件路径* @param sampleRate 采样率，例如 44100* @param channels 声道数 单声道 1 或 双声道 2* @param bitNum 采样位数，8 或 16*/fun convertPcm2Wav(inPcmFilePath: String?,outWavFilePath: String?,sampleRate: Int ,channels: Int ,bitNum: Int) {MLog.d(TAG, "convertPcm2Wav: inPcmFilePath = $inPcmFilePath, outWavFilePath = $outWavFilePath")if (inPcmFilePath.isNullOrEmpty() || outWavFilePath.isNullOrEmpty()) {MLog.d(TAG,"convertPcm2Wav: 文件路径为空!!!")return}MLog.d(TAG, "convertPcm2Wav: 开始转 WAV")var `in`: FileInputStream? = nullvar out: FileOutputStream? = nullval data = ByteArray(1024)try {`in` = FileInputStream(inPcmFilePath)out = FileOutputStream(outWavFilePath)// PCM 文件大小val totalAudioLen = `in`.channel.size()writeWaveFileHeader(out, totalAudioLen, sampleRate, channels, bitNum)var length = 0while (`in`.read(data).also { length = it } > 0) {out.write(data, 0, length)}} catch (e: Exception) {MLog.d(TAG, "convertPcm2Wav: error = $e")e.printStackTrace()} finally {// 编码结束后删除文件 .tempMLog.d(TAG, "convertPcm2Wav: deleteFile = $inPcmFilePath")File(inPcmFilePath).delete()MLog.d(TAG, "convertPcm2Wav: 关闭流")if (`in` != null) {try {`in`.close()} catch (e: IOException) {e.printStackTrace()}}if (out != null) {try {out.close()} catch (e: IOException) {e.printStackTrace()}}}}/*** 输出 WAV 文件* @param out WAV输出文件流* @param totalAudioLen 整个音频 PCM 数据大小* @param sampleRate 采样率* @param channels 声道数* @param bitNum 采样位数*/private fun writeWaveFileHeader(out: FileOutputStream, totalAudioLen: Long,sampleRate: Int, channels: Int, bitNum: Int) {val header = getWaveHeader(totalAudioLen, sampleRate, channels, bitNum)out.write(header, 0, AudioConstant.WAVE_HEAD_SIZE)}/*** 获取Wav header 字节数据* @param totalAudioLen 整个音频PCM数据大小* @param sampleRate 采样率* @param channels 声道数* @param bitNum 采样位数*/fun getWaveHeader(totalAudioLen: Long, sampleRate: Int, channels: Int, bitNum: Int): ByteArray {// 总大小，由于不包括 RIFF 和 WAV，所以是 44 - 8 = 36，在加上 PCM 文件大小val totalDataLen = totalAudioLen + 36// 采样字节 byte 率val byteRate = (sampleRate * channels * bitNum / 8).toLong()val header = ByteArray(AudioConstant.WAVE_HEAD_SIZE)header[0] = 'R'.code.toByte() // RIFFheader[1] = 'I'.code.toByte()header[2] = 'F'.code.toByte()header[3] = 'F'.code.toByte()header[4] = (totalDataLen and 0xffL).toByte() // 数据大小header[5] = (totalDataLen shr 8 and 0xffL).toByte()header[6] = (totalDataLen shr 16 and 0xffL).toByte()header[7] = (totalDataLen shr 24 and 0xffL).toByte()header[8] = 'W'.code.toByte() // WAVEheader[9] = 'A'.code.toByte()header[10] = 'V'.code.toByte()header[11] = 'E'.code.toByte()// FMT Chunkheader[12] = 'f'.code.toByte() // 'fmt 'header[13] = 'm'.code.toByte()header[14] = 't'.code.toByte()header[15] = ' '.code.toByte() // 过渡字节// 数据大小header[16] = 16 // 4 bytes: size of 'fmt ' chunkheader[17] = 0header[18] = 0header[19] = 0// 编码方式 10H 为 PCM 编码格式header[20] = 1 // format = 1header[21] = 0// 通道数header[22] = channels.toByte()header[23] = 0// 采样率，每个通道的播放速度header[24] = (sampleRate and 0xff).toByte()header[25] = (sampleRate shr 8 and 0xff).toByte()header[26] = (sampleRate shr 16 and 0xff).toByte()header[27] = (sampleRate shr 24 and 0xff).toByte()// 音频数据传送速率,采样率*通道数*采样深度/8header[28] = (byteRate and 0xffL).toByte()header[29] = (byteRate shr 8 and 0xffL).toByte()header[30] = (byteRate shr 16 and 0xffL).toByte()header[31] = (byteRate shr 24 and 0xffL).toByte()// 确定系统一次要处理多少个这样字节的数据，确定缓冲区，通道数*采样位数header[32] = (channels * 16 / 8).toByte()header[33] = 0// 每个样本的数据位数header[34] = 16header[35] = 0// Data chunkheader[36] = 'd'.code.toByte() // dataheader[37] = 'a'.code.toByte()header[38] = 't'.code.toByte()header[39] = 'a'.code.toByte()header[40] = (totalAudioLen and 0xffL).toByte()header[41] = (totalAudioLen shr 8 and 0xffL).toByte()header[42] = (totalAudioLen shr 16 and 0xffL).toByte()header[43] = (totalAudioLen shr 24 and 0xffL).toByte()return header}
}
2.2 pcm 裁剪
/*** wav 音频文件截取工具*/
object WavCutUtils {private const val TAG = "WavCutUtils"/* 一个 int 所占字节大小*/private const val INT_BYTE_SIZE = 4/*** 截取 wav 音频文件* @param sourceFile 源文件地址* @param targetFile 目标文件地址* @param start 截取开始时间（秒）* @param end 截取结束时间（秒）* @param duration 总时长（秒）*/fun cutWavAudio(sourceFile: String,targetFile: String,start: Int,end: Int,duration: Int): Boolean {MLog.d(TAG, "cutWavAudio: sourceFile = $sourceFile, targetFile = $targetFile, start = $start, end = $end, duration = $duration")val wav = File(sourceFile)try {val src = RandomAccessFile(wav, "r")val headSize = getHeadSize(src)MLog.d(TAG, "cutWavAudio: headSize = $headSize")if (!wav.exists()) {return false}if (start < 0 || end <= 0 || start >= duration || end > duration || start >= end) {return false}val fis = FileInputStream(wav)// 音频数据大小（wav文件头长度不一定是44）val wavSize = (wav.length() - headSize).toInt()// 截取的音频数据大小val splitSize = wavSize / duration * (end - start)// 截取时跳过的音频数据大小val skipSize = wavSize / duration * startMLog.d(TAG, "cutWavAudio: wavSize = $wavSize, splitSize = $splitSize, skipSize = $skipSize")// 核心操作coreOperate(fis, targetFile, splitSize, headSize, skipSize)} catch (e: Exception) {e.printStackTrace()return false}return true}/*** 核心处理数据* @param fis 输入流* @param targetFile 目标文件路径* @param splitSize 截取的音频数据大小* @param headSize 头部信息大小* @param skipSize 跳过的音频数据大小*/private fun coreOperate(fis: FileInputStream, targetFile: String, splitSize: Int, headSize: Int, skipSize: Int) {// 存放文件大小, 4 代表一个 int 占用字节数val buf1 = ByteBuffer.allocate(INT_BYTE_SIZE)// 放入文件长度信息buf1.putInt(splitSize + 36)// 代表文件长度var fileLen = buf1.array()// 存放音频数据大小，4 代表一个 int 占用字节数val buf2 = ByteBuffer.allocate(INT_BYTE_SIZE)// 放入数据长度信息buf2.putInt(splitSize)// 代表数据长度var dataLen = buf2.array()// 数组反转fileLen = reverse(fileLen)dataLen = reverse(dataLen)val head = ByteArray(headSize)// 读取源 wav 文件头部信息fis.read(head, 0, head.size)// 4 代表一个 int 占用字节数for (i in 0..3) {// 替换原头部信息里的文件长度head[i + INT_BYTE_SIZE] = fileLen[i]// 替换原头部信息里的数据长度head[i + headSize - INT_BYTE_SIZE] = dataLen[i]}// 存放截取的音频数据val fileByte = ByteArray(splitSize + head.size)// 放入修改后的头部信息for (i in head.indices) {fileByte[i] = head[i]}// 存放截取时跳过的音频数据val skipBytes = ByteArray(skipSize)// 跳过不需要截取的数据fis.read(skipBytes, 0, skipBytes.size)// 读取要截取的数据到目标数组fis.read(fileByte, head.size, fileByte.size - head.size)fis.close()// 如果目标文件已存在，则删除目标文件val target = File(targetFile)if (target.exists()) {val result = target.delete()}// 给目标文件写入信息val fos = FileOutputStream(target)fos.write(fileByte)fos.flush()fos.close()}/*** 数组反转* @param array*/private fun reverse(array: ByteArray): ByteArray {var temp: Byteval len = array.sizefor (i in 0 until len / 2) {temp = array[i]array[i] = array[len - 1 - i]array[len - 1 - i] = temp}return array}/*** 得到头文件大小*/private fun getHeadSize(srcFis: RandomAccessFile): Int {var offset = 0// riffgetChunkId(srcFis)offset += 4// lengthgetChunkSize(srcFis)offset += 4// wavegetChunkId(srcFis)offset += 4// fmtgetChunkId(srcFis)offset += 4// fmt lengthvar skipLength = getChunkSize(srcFis)offset += 4var skipBytes = ByteArray(skipLength)srcFis.read(skipBytes)offset += skipLengthvar chunkId = getChunkId(srcFis)offset += 4while (chunkId != "data") {skipLength = getChunkSize(srcFis)offset += 4skipBytes = ByteArray(skipLength)srcFis.read(skipBytes)offset += skipLengthchunkId = getChunkId(srcFis)offset += 4}offset += 4MLog.d(TAG, "getHeadSize: offset = $offset")return offset}/*** 获取块大小*/private fun getChunkSize(srcFis: RandomAccessFile): Int {val formatSize = ByteArray(4)srcFis.read(formatSize)val fisrt8 = formatSize[0].toInt() and 0xFFval fisrduration6 = formatSize[1].toInt() and 0xFFval fisrt24 = formatSize[2].toInt() and 0xFFval fisrt32 = formatSize[3].toInt() and 0xFFval chunkSize = fisrt8 or (fisrduration6 shl 8) or (fisrt24 shl 16) or (fisrt32 shl 24)MLog.d(TAG, "getChunkSize: chunkSize = $chunkSize")return chunkSize}/*** 获取块标识*/private fun getChunkId(srcFis: RandomAccessFile): String {val bytes = ByteArray(4)srcFis.read(bytes)val stringBuilder = StringBuilder()for (i in bytes.indices) {stringBuilder.append(Char(bytes[i].toUShort()))}val chunkId = stringBuilder.toString()MLog.d(TAG, "getChunkId: chunkId = $chunkId")return chunkId}
}
