Audio formats for audiobooks.

[atomotic]

I am starting a new discussion dedicated to audio formats for audiobooks. In previous community calls, we talked about the importance of making format conversion operations available in go-toolkit.

However, before we start coding, it is essential to create a matrix table of the formats we wish to support. This step is crucial because adding dependencies may increase the complexity unnecessarily.

We need to analyze format support in two steps:

Input conversion or recoding

Given an origin directory containing audio files in format {source} we may want to :

• convert them in format {destination} into a temporary directory before packaging them
• recoding (changing bitrate etc) and saving new files into a temporary directory before packaging the

The tool for the job is ffmpeg, ffmpeg-go is quite simple to use. It's just a wrapper around the ffmpeg binary, which should be available in the os running the cli. I wrote about it here:
#71 (reply in thread)

Manifest generation

When generating Readium Manifest we need to extract two properties of the audio file: the mimetype and the duration

mimetype

mimetype can be used for extracting the mimetype

duration

To identify the duration of multiple file formats, we have two choiches:

1. Use a decoder library

Until now, I've only supported mp3 and used and I've used go-mp3 in this way:

audiofile, _ := os.Open(file)
d, _ := mp3.NewDecoder(audiofile)
const sampleSize = 4
samples := d.Length() / sampleSize
audioLength := samples / int64(d.SampleRate())
duration = float64(audioLength)

What other format should we support? We need to use a decoder library for each format and we should prefer a pure Go implementation. Here are some libraries we can use:

• flac
• aac
• opus

2. Use ffprobe

~ ffprobe -show_entries format=filename,size,duration -of json -i natureofacrime_00_conrad.opus 

{
Input #0, ogg, from 'natureofacrime_00_conrad.opus':
  Duration: 00:10:59.42, start: 0.000000, bitrate: 31 kb/s
  Stream #0:0: Audio: opus, 48000 Hz, mono, fltp
    Metadata:
      encoder         : Lavc59.37.100 libopus
      title           : 00 - Preface
      artist          : Joseph Conrad
      track           : 1
      album           : The Nature of a Crime
      genre           : speech
      TLEN            : 659.46
      comment         : https://archive.org/details/nature_crime_ml_2302_librivox
    "format": {
        "filename": "natureofacrime_00_conrad.opus",
        "duration": "659.421458",
        "size": "2628238"
    }
}

go-ffprobe can be used, but likewise go-ffmpeg needs the binary available in the OS.

[mickael-menu]

Regarding mimetype detection, we already have our own utility in the toolkit which should be preferred: https://github.com/readium/go-toolkit/blob/main/pkg/mediatype/mediatype.go

If really needed, we might extend it using third-party dependencies like mimetype.