Audio seg duration

[elv-reza]

The function calcSegDuration() has a bug in setting audio seg duration.

 calcSegDuration({sourceTimescale}) {

    let videoStream = this.getStreamDataForCodecType("video");
    let frameRate = videoStream.frame_rate;

    let seg ={};
    switch (frameRate) {
      case "24":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 48;
        seg.duration = "30";
        break;
      case "25":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 50;
        seg.duration = "30";
        break;
      case "30":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 60;
        seg.duration = "30";
        break;
      case "30000/1001":
        seg.video = 30.03 * sourceTimescale;
        seg.audio = 29.76 * 48000;
        seg.keyint = 60;
        seg.duration = "30.03";
        break;
      case "48":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 96;
        seg.duration = "30";
        break;
      case "50":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 100;
        seg.duration = "30";
        break;
      case "60":
        seg.video = 30 * sourceTimescale;
        seg.audio = 30 * 48000;
        seg.keyint = 120;
        seg.duration = "30";
        break;
      case "60000/1001":
        seg.video = 30.03 * sourceTimescale;
        seg.audio = 29.76 * 48000;
        seg.keyint = 120;
        seg.duration = "30.03";
        break;
      default:
        console.log("Unsupported frame rate", frameRate);
        break;
    }
    return seg;
  }

The function can simply set the audio seg duration to 29.76 * 48000 for all the cases as follows. Notice that the audio seg duration does not have dependency to video frame rate at all and must not be determined by that.

 calcSegDuration({sourceTimescale}) {

    let videoStream = this.getStreamDataForCodecType("video");
    let frameRate = videoStream.frame_rate;

    let seg ={};
    seg.audio = 29.76 * 48000;

    switch (frameRate) {
      case "24":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 48;
        seg.duration = "30";
        break;
      case "25":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 50;
        seg.duration = "30";
        break;
      case "30":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 60;
        seg.duration = "30";
        break;
      case "30000/1001":
        seg.video = 30.03 * sourceTimescale;
        seg.keyint = 60;
        seg.duration = "30.03";
        break;
      case "48":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 96;
        seg.duration = "30";
        break;
      case "50":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 100;
        seg.duration = "30";
        break;
      case "60":
        seg.video = 30 * sourceTimescale;
        seg.keyint = 120;
        seg.duration = "30";
        break;
      case "60000/1001":
        seg.video = 30.03 * sourceTimescale;
        seg.keyint = 120;
        seg.duration = "30.03";
        break;
      default:
        console.log("Unsupported frame rate", frameRate);
        break;
    }
    return seg;
  }

linke to the code:

elv-live-js/src/LiveConf.js

Line 50 in ba848d4

calcSegDuration({sourceTimescale}) {

[elv-gerald]

added these changes with #154

[elv-reza]

@elv-gerald, @elv-zenia The above solution that I gave is very simple and it covers most of the use cases that we have. But to cover all the cases we need more work to be done.

• When you probe the input stream you obtain input_sample_rate for audio. If the input audio is AAC and input_sample_rate is in {8000, 12000, 16000, 22050, 24000, 32000, 44100, 48000, 88200, 96000} then we use the input_sample_rate and set sample_rate = input_sample_rate.
• Otherwise we set sample_rate = 48000
• Now audio seg duration is calculated by: audio_seg_duration = 29.76 * sample_rate
• Notice that input stream might have multiple audio streams, in above input_sample_rate refers to sample rate of selected audio stream (i.e index 1 or 2 ...).

This should cover all the cases and we need to test it with many different input streams (close the issue after testing).

[elv-gerald]

Does the above apply to both udp and rtmp?

[elv-reza]

Applies to both.

[elv-zenia]

Will make other necessary changes but for now, updated the live stream app to use the fixed value 29.76 * 48000. Deployed on demo, prod-dev, and prod.