half-speed play with no audio pitch change? (almost clueless)

[gregconquest]

I have a particular video, "OH! Mikey" (wikipedia (http://en.wikipedia.org/wiki/The_Fuccons)), that has very fast spoken Japanese with English subtitles. I want to play it at half-speed -- the video and the audio both slowed down, with no pitch change, so I can read the subtitles and understand what is being said. I remember WinDVD or PowerDVD had this feature years ago, but I cannot find out much about it on any player now. Either this control is a specialized feature that is not widely supported, or I'm using the wrong search terms, "slow motion", "pitch change", . . .

I originally asked this question here at videohelp.com (http://forum.videohelp.com/topic350432.html?sid=fa86cfc617dc526b47884dc797ce0775#1845630), but they advised me to try doom9 and perhaps avisynth . . .

I am mostly using linux (ubuntu), but I do have a Windows XP PC available as well. I don't have Vegas Video, Premiere, or any other big commercial video editing apps installed now.

So, can anyone here give me suggestions as to how to play this video at other speeds? And since it is mostly this one video I want to play slow, is there a way to transcode it to always play slower in any player -- to basically double the file size rendering every video frame twice while doing something analogous to the audio?

Thanks for any help,
Greg Conquest

keys: linux fast audio speak quickly too quickly speaking

[IanB]

:search:...
SelectEvery(1, 0, 0).AssumeFPS(Last) # Cheap 2x slow mo
TimeStretch(Tempo=50.0) # Audio half speed, same pitch

[gregconquest]

Thanks for this, IanB, and I'm sorry to be so slow in following up on my question. I hate loose threads . . .

As I'm mostly running Linux, I have not installed avisynth. It has been on my list of things to do, but I now have a different solution to my problem. I used avidemux and audacity to slow down the video.

Here are my steps (I do still have to use Windows for part of this, but that issue will resolve itself soon):

1) Open video in avidemux
2) Audio -- Save (as mp3...)
2b)use audacity to stretch audio
(change tempo, no pitch change... one hour processing for 66minutes of mp3 on my 3GHz system (tempo change not possible in Linux version as of summer 2008))
3) Change Video from "Copy" to matching (MPEG-4 ASP (Xvid4). Match as close as possible, in my case)
4) Check "Configure". Check "Two-pass"...
5) "Filter" Resample fps (from 29.970 to 59.94 in my case)
6) Audio -- don't worry. We will dispose of it. avidemux cannot change it freely...
7) Format -- make your choice (MP4)
8) Save (give incremental name -- OhMikey1.2.doublefps.mp4, in my case)
9) Close and Open file just created (OhMikey1.2.doublefps.mp4)
10) Video -- Frame Rate -- cut in half (59.940 to 29.970)
11) Audio -- select stretched audio from 2b).
12) Save.

Now you'll have your original video stretched out to double its length.

I have found one seemingly intractable problem. The audio gets warbled in the process of stretching it out. I thought maybe audacity was just doing a poor job, so I downloaded a sine wave file and tried stretching it. It was perfect -- no artefacts.

I think what is happening is the result of changing pitches. I haven't fully followed up on this, but if I sing slowly changing pitch and then use audacity to stretch out the audio, I think it would be fine. However, if I go through rapid pitch changes, the sound samples that are being duplicated are going to produce warbles. They are kind of like audio stair steps, best I can figure. The only way to overcome this would be to use higher frequency sound samples, I mean samples per second, as in MD at 32KHz, CD at 44.1 KHz, DAT at 48 KHz, etc. Smaller stair steps would equal less warbling. But I am not sure of my analysis. There is a rapidly rising whistle sound at the beginning of both clips below which seems to not suffer from the warbling. The warbling/tremelo wavering is prominent in human voices, though.

Here is a sample of what I did (I presume no one is going to protest me posting these links. They are both small samples of the original. The content is incidental to the discussion at hand. It seems quite clearly to lie within the realm of fair use...):

Cathy at regular speed
http://uk.youtube.com/watch?v=NRfvIDdWeQo

Cathy 1 half-speed
http://uk.youtube.com/watch?v=5_Knx4Z7rgE

Greg

[Gavino]

Why not give the Avisynth solution a try? It only takes a minute to install, and for the purposes of this forum it would be useful to know how the built-in TimeStretch compares to Audacity.

Bear in mind that no audio processing algorithm is perfect, and something like this requires the best quality source material you can give it. What is the audio format and sampling rate of your original video? If you can save and process it as WAV rather than MP3 you will likely see an improvement, as MP3 (and all lossy compression) introduces artefacts that are frequency-dependent.

[IanB]

@gregconquest,

Avisynth apparently works quite will under Wine, you just need a suitable host application to run it with, possible just avs2wav in this case.

IIRC audacity uses the same SoundTouch library (Copyright © Olli Parviainen 2002-2006) as TimeStretch does.

This tempo changing algorithm works by chopping the audio stream into short segments then repeating and pasting them back together such that the correlation at the join is maximal. The default parameters Olli implemented are Sequence = 82ms, SeekWindow = 28ms and Overlap = 12ms.... go through rapid pitch changes, the sound samples that are being duplicated are going to produce warbles ...The warble effect is caused by the repeating of the slices when there is appreciable change in the sound during the sequence slices. Olli recommends increasing the Sequence value to combat this effect when slowing the tempo. I have found you also need to increase the SeekWindow and Overlap values in roughly the same proportions to avoid a hunting or breathing effect. You may need to process different parts of the audio stream with different settings to achieve an optimal result. Music, speech and singing all have different requirements for successful use with this algorithm.

Olli has standalone applications using his libraries, have a look at http://www.iki.fi/oparviai/soundtouch or http://www.surina.net/soundtouch for downloads and further documentation.

You might also look into BeSweet for doing your audio stretch.


And you can probably short circuit a whole lot of processing by just changing the video rate in the header to 14.985 and muxing in the reprocessed audio, i.e. steps 1, 2, 2b & 10, 11, 12.