Recherche avancée

Sur d’autres sites (3504)

• bad audio mic recording quality with ffmpeg compared to sox

  1er juillet 2021, par user2355330

  I am contacting you as after 3 days of searching I am stuck on a really simple point.

  


  I want to record the sound of my mic on MacOS using ffmpeg.

  


  I managed to do it using the following command :

  


  ffmpeg -f avfoundation -audio_device_index 2 -i "none:-" -c:a pcm_s32l alexspeaking.wav -y -loglevel debug


  


  The issue is that each time I am speaking, there are cracks and pop in the sound...

  


  I tried to use sox and it gave me a perfect and crystal clear sound and I have no idea why... Below is the output of the sox command :

  


  sox -t coreaudio "G935 Gaming Headset" toto.wav -V6
sox:      SoX v
time:     Nov 15 2020 01:06:02
uname:    Darwin MacBook-Pro.local 20.5.0 Darwin Kernel Version 20.5.0: Sat May  8 05:10:33 PDT 2021; root:xnu-7195.121.3~9/RELEASE_X86_64 x86_64
compiler: gcc Apple LLVM 12.0.0 (clang-1200.0.32.27)
arch:     1288 48 88 L
sox INFO coreaudio: Found Audio Device "DELL U2721DE"
sox INFO coreaudio: Found Audio Device "G935 Gaming "
sox DBUG coreaudio: audio device did not accept 2 channels. Use 1 channels instead.
sox DBUG coreaudio: audio device did not accept 44100 sample rate. Use 48000 instead.
Input File     : 'G935 Gaming Headset' (coreaudio)
Channels       : 1
Sample Rate    : 48000
Precision      : 32-bit
Sample Encoding: 32-bit Signed Integer PCM
Endian Type    : little
Reverse Nibbles: no
Reverse Bits   : no
sox INFO sox: Overwriting `toto.wav'
sox DBUG wav: Writing Wave file: Microsoft PCM format, 1 channel, 48000 samp/sec
sox DBUG wav:         192000 byte/sec, 4 block align, 32 bits/samp
Output File    : 'toto.wav'
Channels       : 1
Sample Rate    : 48000
Precision      : 32-bit
Sample Encoding: 32-bit Signed Integer PCM
Endian Type    : little
Reverse Nibbles: no
Reverse Bits   : no
Comment        : 'Processed by SoX'
sox DBUG effects: sox_add_effect: extending effects table, new size = 8
sox INFO sox: effects chain: input        48000Hz  1 channels (multi) 32 bits unknown length
sox INFO sox: effects chain: output       48000Hz  1 channels (multi) 32 bits unknown length
sox DBUG sox: start-up time = 0.051332
In:0.00% 00:00:07.13 [00:00:00.00] Out:340k  [      |      ]        Clip:0    ^C
sox DBUG input: output buffer still held 2048 samples; dropped.
Aborted.
sox DBUG wav: Finished writing Wave file, 1359872 data bytes 339968 samples


  


  I am pretty sure the issue is linked to the way the encoding is done and the params I used with ffmpeg but I don't seem to be able to grasp which one I must use.

  


  Any ideas if there are ffmpeg experts here ?

  


• how to automatically oversample and downsample when using filters ?

  4 juin 2021, par Yue Wang

  I am researching avfilters. Say I have an audio with sample rate s, and bit depth d,

  


  What I want to do is to write a graph that

  


  

  • upsample s by 4x, and set precision to 64bit float

  


  • apply some biquad filters with 64bit precision

  


  • downsample by 4x back to s, and set bit depth back to d.

  


  


  The reason to oversample is to get better filtering result by antialiasing, and the reason to downsample is to stream using the original source format.

  


  I don't know if there's way that I can do it automatically in the graph.

  


  ashowinfo could print out the sample rate, but seems there's no way to use the value later in the pipeline.

  


  asoftclip has a oversample factor. but it's not available in other filters.

  


• ffmpeg - pts drift after seek with codec copy

  24 mars 2021, par Andrey Rikunov

  There are two ffmpeg commands. First one is used to seek and copy video chunk. Second one is used to transcode video chunk applying select filter for exact frames match.

  


  Here is how :

  


  

  1. ffmpeg -ss  -to  -copyts -i <input /> -map 0:v:0 -c copy chunk.mp4

  


  2. ffmpeg -copyts -i chunk.mp4 -vf 'select=between(pts\,\,)' transcoded_cunk.mp4

  


  


  It works fine most of the times. But for some inputs there is a little pts drift in downloaded chunk so missing frames is possible. In other words pts of the same packets (compared by hash) are shifted by several points (in my case 0,0002 sec) between input and chunked output.

  


  What is the possible reason for such pts drift ?

  


  UPDATE 1 : That's because ffmpeg set timescale=1000 in mvhd atom so edit list media time to start from looses precision. Is it possible to force mvhd timescale ?

  


  UPDATE 2 : It's not possible to change mvhd timescale because ffmpeg uses constant (MOV_TIMESCALE 1000) :

  


  https://github.com/FFmpeg/FFmpeg/blob/82bd02a2c73bb5e6b7cf5e5eba486e279f1a7358/libavformat/movenc.c#L3498

  


  UPDATE 3 : same issue discussed earlier