1

i'm wanting to get the duration of .caf audio files using go. I found a few decoders but their Duration() methods just return 0 with comments perhaps suggesting ways of calculating the duration, does any know if these comments are legit and if so, how I might calculate the duration? I'll accept "it's not possible" as an answer if there's no easy solution.

func (d *Decoder) Duration() time.Duration {
    //duration := time.Duration((float64(p.Size) / float64(p.AvgBytesPerSec)) * float64(time.Second))
    //duration := time.Duration(float64(p.NumSampleFrames) / float64(p.SampleRate) * float64(time.Second))

    return 0
}

one implementation example although i'm happy to use any implementation that's easy to install: https://github.com/mattetti/audio/blob/master/caf/decoder.go

Rusty Rob
  • 16,489
  • 8
  • 100
  • 116
  • here's the implementation: https://github.com/mattetti/audio/blob/master/caf/decoder.go – Rusty Rob May 11 '20 at 05:51
  • We can't guess why those lines are commented out, but have you tried them and compared the result to the actual duration? – Marc May 13 '20 at 16:38

1 Answers1

1

The doc comments in that file you linked is taken directly from Apple's spec. In those docs, you'll find these two important things:

"The duration of the audio in the file is [the number of valid frames] divided by the sample rate specified in the file’s Audio Description chunk."

OK, cool, but how many valid frames are there? There are two possible ways to know:

  • If the CAF has a packet table, it must include the number of valid frames. Perfect.
  • The only CAFs that are allowed to NOT have a packet table are those with constant packet sizes:

"Note that, as long as the format has a constant number of frames per packet, you can calculate the duration of each packet by dividing the mSampleRate [frames per second] value by the mFramesPerPacket value."

That tells you the duration per packet, but because packets are a constant size, the number of packets is just the audioDataSize / bytesPerPacket. The latter value is included in the Audio Description. The former is often embedded directly into the file, but it's permitted to be -1 with the audio data as the last chunk, in which case its size is totalFileSize - startOfAudioData

It breaks down like this:

  • If there's a Packet Table Chunk, use it and the Audio Description: seconds = validFrames / sampleRate
  • Otherwise, packets must have constant size:
    • framesPerByte = framesPerPacket / bytesPerPacket
    • seconds = framesPerByte * audioDataSize

The library you've got reads the Audio Description chunk, but I don't think it reads the Packet Table. Also, I'm not confident it calculates the audio data size if the chunk is -1. Maybe it does both/either, in which case, you can use the information above.

If not, you can just parse the file yourself, especially if you only care about the duration. The file starts with a short header, then is split into "chunks" (aka TLVs). Here's a sample implementation you can use as a starting point or to modify the library you linked:



func readCAF() { 
    buf := []byte{
        // file header
        'c', 'a', 'f', 'f', // file type
        0x0, 0x1, 0x0, 0x0, // file version, flags

        // audio description
        'd', 'e', 's', 'c', // chunk type
        0x0, 0x0, 0x0, 0x0,
        0x0, 0x0, 0x0, 0x20, // CAFAudioFormat size

        0x40, 0xe5, 0x88, 0x80,
        0x00, 0x00, 0x00, 0x00, // sample rate
        'l', 'p', 'c', 'm', // fmt id
        0x0, 0x0, 0x0, 0x0, // fmt flags
        0x0, 0x0, 0x0, 0x1, // bytes per packet
        0x0, 0x0, 0x0, 0x1, // frames per packet
        0x0, 0x0, 0x0, 0x2, // channels per frame
        0x0, 0x0, 0x0, 0x3, // bits per channel

        // audio data
        'd', 'a', 't', 'a', // chunk type
        0xff, 0xff, 0xff, 0xff,
        0xff, 0xff, 0xff, 0xff, // size of data section (-1 = til EOF)

        // actual audio packets (in theory, anyway)
        0x0,
        0x0,
        0x0,
        0x0,
        0x0,
        0x0,
    }

    fileSize := len(buf)
    br := bufio.NewReader(bytes.NewBuffer(buf))

    type cafHdr struct {
        Typ     [4]byte
        Version uint16
        _       uint16
    }

    type chunkHdr struct {
        Typ [4]byte
        Sz  int64
    }

    type audioDescription struct {
        FramesPerSec     float64
        FmtId            uint32
        FmtFlags         uint32
        BytesPerPacket   uint32
        FramesPerPacket  uint32
        ChannelsPerFrame uint32
        BitsPerChannel   uint32
    }

    type packetTable struct {
        NPackets, NValidFrames, NPrimingFr, NRemainingFr int64
    }

    const FileHeaderSz = 8
    const ChunkHeaderSz = 12
    const AudioDescSz = 32
    const PacketHdrSz = 24

    fileHdr := cafHdr{}
    if err := binary.Read(br, binary.BigEndian, &fileHdr); err != nil {
        panic(err)
    }
    if fileHdr.Typ != [4]byte{'c', 'a', 'f', 'f'} || fileHdr.Version != 1 {
        panic("unknown file format")
    }
    remaining := int64(fileSize) - FileHeaderSz

    audioDesc := audioDescription{}
    packetTab := packetTable{}
    var audioDataSz int64

readChunks:
    for {
        hdr := chunkHdr{}
        if err := binary.Read(br, binary.BigEndian, &hdr); err != nil {
            panic(err)
        }
        remaining -= ChunkHeaderSz

        switch hdr.Typ {
        case [4]byte{'d', 'e', 's', 'c'}: // audio description
            if err := binary.Read(br, binary.BigEndian, &audioDesc); err != nil {
                panic(err)
            }
            hdr.Sz -= AudioDescSz
            remaining -= AudioDescSz

        case [4]byte{'p', 'a', 'k', 't'}: // packet table
            if err := binary.Read(br, binary.BigEndian, &packetTab); err != nil {
                panic(err)
            }
            hdr.Sz -= PacketHdrSz
            remaining -= PacketHdrSz

        case [4]byte{'d', 'a', 't', 'a'}: // audio data
            if hdr.Sz > 0 {
                audioDataSz = hdr.Sz
            } else if hdr.Sz == -1 {
                // if needed, read to EOF to determine byte size
                audioDataSz = remaining
                break readChunks
            }
        }

        if hdr.Sz < 0 {
            panic("invalid header size")
        }
        remaining -= hdr.Sz

        // Skip to the next chunk. On 32 bit machines, Sz can overflow,
        // so you should check for that (or use Seek if you're reading a file).
        if n, err := br.Discard(int(hdr.Sz)); err != nil {
            if err == io.EOF && int64(n) == hdr.Sz {
                break
            }
            panic(err)
        }
    }

    var seconds float64

    // If the data included a packet table, the frames determines duration.
    if packetTab.NValidFrames > 0 {
        seconds = float64(packetTab.NValidFrames) / audioDesc.FramesPerSec
    } else {
        // If there no packet table, it must have a constant packet size.
        if audioDesc.BytesPerPacket == 0 || audioDesc.FramesPerPacket == 0 {
            panic("bad data")
        }
        framesPerByte := float64(audioDesc.FramesPerPacket) / float64(audioDesc.BytesPerPacket)
        seconds = framesPerByte * float64(audioDataSz)
    }

    fmt.Printf("seconds: %f\n", seconds)
}

Saites
  • 792
  • 7
  • 11