Recherche avancée

Sur d’autres sites (7742)

• ffmpeg to count words in audio text

  17 juillet 2020, par Joel Parker

  I am new to signal processing but wanted to take an audio file and determine how many words are spoken in one minute. I was thinking I could use the top of the loudness peaks to count the words but do not quite understand how to achieve this.

  


  First I used ffmpeg to remove the audio from the mp4 file I am using :

  


  ffmpeg -i courtcase.mp4 audiofile.mp4

  


  Then I tried to detect the loudness :

  


  ffmpeg -t 10 -i audiofile.mp4 -af "volumedetect" -f null /dev/null

  


  This produced some statistical information :

  


  video:157kB audio:1723kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] n_samples: 882000
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] mean_volume: -20.6 dB
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] max_volume: -4.0 dB
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] histogram_4db: 64
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] histogram_5db: 88
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] histogram_6db: 220
[Parsed_volumedetect_0 @ 0x7fa6b26068c0] histogram_7db: 843



  


  I am not sure why it still shows 157kB of video, maybe my first command is wrong ?

  


  Anyway, assuming the file is just audio I found this command, which I believe shows dbm slices for 10 seconds :

  


  ffmpeg -i audiofile.mp4 -af astats=metadata=1:reset=1,ametadata=print:key=lavfi.astats.Overall.RMS_level:file=- -f null -


  


  and it produced a bunch of output :

  


  video:5782kB audio:63504kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: unknown
[Parsed_astats_0 @ 0x7ff74c004bc0] Channel: 1
[Parsed_astats_0 @ 0x7ff74c004bc0] DC offset: 0.000240
[Parsed_astats_0 @ 0x7ff74c004bc0] Min level: -0.166239
[Parsed_astats_0 @ 0x7ff74c004bc0] Max level: 0.127112
[Parsed_astats_0 @ 0x7ff74c004bc0] Min difference: 0.000003
[Parsed_astats_0 @ 0x7ff74c004bc0] Max difference: 0.025335
[Parsed_astats_0 @ 0x7ff74c004bc0] Mean difference: 0.004455
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS difference: 0.006165
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak level dB: -15.585332
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS level dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS peak dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS trough dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] Crest factor: 3.414311
[Parsed_astats_0 @ 0x7ff74c004bc0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak count: 2
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor dB: nan
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor count: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Bit depth: 32/32
[Parsed_astats_0 @ 0x7ff74c004bc0] Dynamic range: 72.297593
[Parsed_astats_0 @ 0x7ff74c004bc0] Zero crossings: 74
[Parsed_astats_0 @ 0x7ff74c004bc0] Zero crossings rate: 0.072266
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of NaNs: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of Infs: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of denormals: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Channel: 2
[Parsed_astats_0 @ 0x7ff74c004bc0] DC offset: 0.000240
[Parsed_astats_0 @ 0x7ff74c004bc0] Min level: -0.166239
[Parsed_astats_0 @ 0x7ff74c004bc0] Max level: 0.127112
[Parsed_astats_0 @ 0x7ff74c004bc0] Min difference: 0.000003
[Parsed_astats_0 @ 0x7ff74c004bc0] Max difference: 0.025335
[Parsed_astats_0 @ 0x7ff74c004bc0] Mean difference: 0.004455
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS difference: 0.006165
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak level dB: -15.585332
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS level dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS peak dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS trough dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] Crest factor: 3.414311
[Parsed_astats_0 @ 0x7ff74c004bc0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak count: 2
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor dB: nan
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor count: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Bit depth: 32/32
[Parsed_astats_0 @ 0x7ff74c004bc0] Dynamic range: 72.297593
[Parsed_astats_0 @ 0x7ff74c004bc0] Zero crossings: 74
[Parsed_astats_0 @ 0x7ff74c004bc0] Zero crossings rate: 0.072266
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of NaNs: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of Infs: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of denormals: 0
[Parsed_astats_0 @ 0x7ff74c004bc0] Overall
[Parsed_astats_0 @ 0x7ff74c004bc0] DC offset: 0.000240
[Parsed_astats_0 @ 0x7ff74c004bc0] Min level: -0.166239
[Parsed_astats_0 @ 0x7ff74c004bc0] Max level: 0.127112
[Parsed_astats_0 @ 0x7ff74c004bc0] Min difference: 0.000003
[Parsed_astats_0 @ 0x7ff74c004bc0] Max difference: 0.025335
[Parsed_astats_0 @ 0x7ff74c004bc0] Mean difference: 0.004455
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS difference: 0.006165
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak level dB: -15.585332
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS level dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS peak dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] RMS trough dB: -26.251394
[Parsed_astats_0 @ 0x7ff74c004bc0] Flat factor: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Peak count: 2.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor dB: nan
[Parsed_astats_0 @ 0x7ff74c004bc0] Noise floor count: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Bit depth: 32/32
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of samples: 1024
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of NaNs: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of Infs: 0.000000
[Parsed_astats_0 @ 0x7ff74c004bc0] Number of denormals: 0.000000
ts_time:368.268
lavfi.astats.Overall.RMS_level=-29.670653
frame:15861 pts:16241664 pts_time:368.292
lavfi.astats.Overall.RMS_level=-30.851195
frame:15862 pts:16242688 pts_time:368.315
lavfi.astats.Overall.RMS_level=-30.700943
frame:15863 pts:16243712 pts_time:368.338
lavfi.astats.Overall.RMS_level=-33.638604
frame:15864 pts:16244736 pts_time:368.361
lavfi.astats.Overall.RMS_level=-21.873170
frame:15865 pts:16245760 pts_time:368.385
lavfi.astats.Overall.RMS_level=-20.001936
frame:15866 pts:16246784 pts_time:368.408
lavfi.astats.Overall.RMS_level=-18.571318
frame:15867 pts:16247808 pts_time:368.431
lavfi.astats.Overall.RMS_level=-18.470749
frame:15868 pts:16248832 pts_time:368.454
lavfi.astats.Overall.RMS_level=-19.506688
frame:15869 pts:16249856 pts_time:368.477
lavfi.astats.Overall.RMS_level=-21.270579
frame:15870 pts:16250880 pts_time:368.501
lavfi.astats.Overall.RMS_level=-25.007862
frame:15871 pts:16251904 pts_time:368.524
lavfi.astats.Overall.RMS_level=-25.654372
frame:15872 pts:16252928 pts_time:368.547
lavfi.astats.Overall.RMS_level=-24.948357
frame:15873 pts:16253952 pts_time:368.57
lavfi.astats.Overall.RMS_level=-30.523540
frame:15874 pts:16254976 pts_time:368.594
....


  


  This is where I'm stuck. I think I have the information I need to determine the number of words spoken in a minute, except I don't know how to put all together. Also the last command just measures 10s slices, would I need to change that to 60s ? Does anyone know how to do this or if there is a better approach ?

  


• How the ffmpeg astats crest factor is calculated

  30 août 2017, par FranGar

  I’m scripting a ffmpeg chain process for my work. The aim is normalizing/compressing lot of audio files (mp3’s).
  It’s done in Python and the critical part is the line :

  ffmpeg -y -i "Input.mp3" -codec:a libmp3lame -b:a 96k -af acompressor=threshold=-15dB:ratio=5:attack=0.01:release=1000:knee=2,dynaudnorm=g=3:m=2:p=0.95 "Output.mp3"

  The python script it’s complete and working BUT the nature of the audios (voice recordings) are very different so I can’t use the same params for all of them.

  I make some experimenting with the values of the ffmpeg filter astats and i discovered that the crest factor (Standard ratio of peak to RMS level ) gave a good reference to programatically get the better params.

  In fact I saw that a recording with a nice dynamic range sound and smooth in shape, get crest values around 9-15 (the compress/normlz params will be somehow conservative). But audios with crest around 22-30 need more aggressive processing.
  (All empirically)

  Somebody can clarify how the crest values are really calculated ? Which are the peaks taken to account ? (Why the flat factor is always 0 ?)
  Or if somebody knows how to get a value representing the sound ’smoothness’ will be nice also.

  Thanks for the ideas.

• How the ffmpeg astats crest factor value of an audio track is calculated

  29 août 2017, par FranGar

  I’m scripting a ffmpeg chain process for my work. The aim is normalizing/compressing lot of audio files (mp3’s).
  It’s done in Python and the critical part is the line :

  ffmpeg -y -i "Input.mp3" -codec:a libmp3lame -b:a 96k -af acompressor=threshold=-15dB:ratio=5:attack=0.01:release=1000:knee=2,dynaudnorm=g=3:m=2:p=0.95 "Output.mp3"

  The python script it’s complete and working BUT the nature of the audios (voice recordings) are very different so I can’t use the same params for all of them.

  I make some experimenting with the values of the ffmpeg filter astats and i discovered that the crest factor (Standard ratio of peak to RMS level ) gave a good reference to programatically get the better params.

  In fact I saw that a recording with a nice dynamic range sound and smooth in shape, get crest values around 9-15 (the compress/normlz params will be somehow conservative). But audios with crest around 22-30 need more aggressive processing.
  (All empirically)

  Somebody can clarify how the crest values are really calculated ? Which are the peaks taken to account ? (Why the flat factor is always 0 ?)
  Or if somebody knows how to get a value representing the sound ’smoothness’ will be nice also.

  Thanks for the ideas.