Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Help improving voice audio recordings (Read 5599 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Help improving voice audio recordings

Hi,
I have many voice recordings that have a pretty bad quality, I would like to improve.
I actually know it is possible to significantly improve, since I heard such a sample, but I don't know how they did it.
I tried to play with many filters in GoldWave, but without success.
Someone's help would be much appreciated!
Below a link to a sample of such file.
http://www.sendspace.com/file/wyo953
Thank you very much!
ML

Help improving voice audio recordings

Reply #1
Yet another download link that will not be used by many because one has to download and install software to use it.

I downloaded the 774 KiB MP3 file directly, without any special software.


On clicking the 4th item mentioning download on the page I finally bypassed the download.

The file itself appears to have a ton of nonlinear distortion but isn't visibly clipped.

That would be a real puzzle!

Suggest starting over with a new recording.

As an aside I've been making recordings using my Sansa Clip+ and Fuze. Several have been of meetings of maybe a dozen people in a medium sized room. Several more were of talks given in rooms with literally thousands of people, and from seats near the back of the room. Not hifi, but amazingly clear and useful.

Help improving voice audio recordings

Reply #2
Yet another download link that will not be used by many because one has to download and install software to use it.


Those huge Download Now or Play Now software ads are such a huge turnoff, hiding the direct download from Sendspace link that skamp found. In my experience, mediafire.com is a little friendlier.

I opened it first in Audacity. It's a 32kbps CBR MP3 at 22.05 kHz sampling rate.

First thing I notice on spectrum view is there's a sharp 4kHz cut-off with a few sharp lines above 4kHz in the spectrogram, possible caused by clipping or aliasing in previous processing. The cut-off implies the original was sampled at 8kHz (e.g. telephone, possibly A-law or mu-law). Some parts sound like roughness caused by a naive (e.g. staircase) sample rate conversion without filtering in the previous processing chain, though the 4kHz cut-off seems pretty sharp. There's a possibility that some voice codec was used, but I can't be sure, especially as I don't understand the language.

On waveform view (Show Clipping) there's some clipping on mp3 decode in much the same places, so I used foobar2000's ReplayGain scanner and Apply Gain to MP3 Data to make the level sensible.

Now with no MP3 decode clipping there are still subtle vertical lines in the spectrogram going as high as about 8 to 8.3 kHz in places. It's possible that clipping occurred during a previous sample rate conversion and it's possible that even a 16kHz sample rate was once used or that the 4-8kHz band is an aliased mirror of the 0-4kHz band in those places. Plenty of places with content only to 4kHz still sound pretty rough.

As a quick operation to remove frequencies above 4kHz, I used fb2k to Resample to 8000 Hz in Ultra Mode then back to 22050 Hz (though going back to 22050 is pointless). Possible minor improvement on vocal plosives etc, but subtle at best.

I tried improving voice clarity by using the Equalizer. The AM Radio preset cuts out the low bass and the treble and focuses on the typical vocal frequencies, for example, but still sounds pretty bad.

Basically, I think theres a lot of distortion baked into your audio, quite possibly impinging on the same frequencies as the speech content, which isn't easy to reverse. I'd love to know if there's anything to help but I don't know of anything myself.
Dynamic – the artist formerly known as DickD

Help improving voice audio recordings

Reply #3
Just as a "reference point"...  Professional recordings are still done by skilled recording engineers,  usually in soundproof studios with good acoustics and with really good equipment, plus good talant, and good instruments when recording music.    No matter how good your software is, there is only so much you can do if you don't start-out with a good recording. 

On-location movie dialog is re-recorded in the studio.  Of course, The Foley (sound effects) and music you hear in movies is done in the studio too.    Radio & TV studios are soundproof, and they use good equipment.    On-location news often has background noise, but again they are using good equipment and good techniques to get the best sound possible.

You can get very good recordings "live" or "at home", but the idea is to start-out with the best recording possible.

Help improving voice audio recordings

Reply #4
Those huge Download Now or Play Now software ads are such a huge turnoff, hiding the direct download from Sendspace link that skamp found. In my experience, mediafire.com is a little friendlier.

I opened it first in Audacity. It's a 32kbps CBR MP3 at 22.05 kHz sampling rate.

First thing I notice on spectrum view is there's a sharp 4kHz cut-off with a few sharp lines above 4kHz in the spectrogram, possible caused by clipping or aliasing in previous processing.
<smip>


Just as a comparison, Mark Levins podcast download is also 16 bit mono, 32 Kbit data rate and sample rate of 22 KHz - identical to this file. The Levin upper cutoff is 7.6 KHz with no upper spectrum glitches. While no where near 'hi fi' it does demonstrate what is possible within these constraints.



Help improving voice audio recordings

Reply #5
Dear friends,
First of all - I'm really impressed by your replies and professionalism. Thank you!
Now, it appears that someone did succeed in improving such audio.
For some of the recordings I found also a better version. (not for all of them, otherwise I wouldn't bother trying to improve myself).
See in below zip 2 files of exactly the same part, one with bad audio the other with better (not excellent though) audio quality.
http://www.sendspace.com/file/xdnmed [this is safe. only click "Click here to start download from sendspace"].
I would really appreciate if someone would be able to advise how to get to a better quality.
Thank you so much.
ML

Help improving voice audio recordings

Reply #6
Dear friends,
First of all - I'm really impressed by your replies and professionalism. Thank you!
Now, it appears that someone did succeed in improving such audio.
For some of the recordings I found also a better version. (not for all of them, otherwise I wouldn't bother trying to improve myself).
See in below zip 2 files of exactly the same part, one with bad audio the other with better (not excellent though) audio quality.
http://www.sendspace.com/file/xdnmed [this is safe. only click "Click here to start download from sendspace"].
I would really appreciate if someone would be able to advise how to get to a better quality.


I really can't say that I think one sounds better than the other.

Knowing nothing about how these files are made, it is competely impossible to suggest any improvements on the methodology.


Help improving voice audio recordings

Reply #8
You can go far with EQ, I suggest reading up on basic parametric EQ operations. Common operations include hipass and lowpass filters to remove rumble and noise respectively. Parametric filters to reduce exagerated ranges in the mids, with care obviously as this is where the majority or the significant info lies. Also a using shelf EQs as for broad titling can do wonders when a recording is generally murky or thin sounding.

I highly recommend using foobar2000 with the VST component and a good free EQ plugin such a MEqualizer, that way you can play around with filters without making any irreversible changes, go back and forth between presets, even export presets to a file which can be conveniently placed in the folder of the recording.

Help improving voice audio recordings

Reply #9
You can go far with EQ, I suggest reading up on basic parametric EQ operations. Common operations include hipass and lowpass filters to remove rumble and noise respectively. Parametric filters to reduce exagerated ranges in the mids, with care obviously as this is where the majority or the significant info lies. Also a using shelf EQs as for broad titling can do wonders when a recording is generally murky or thin sounding.

I highly recommend using foobar2000 with the VST component and a good free EQ plugin such a MEqualizer, that way you can play around with filters without making any irreversible changes, go back and forth between presets, even export presets to a file which can be conveniently placed in the folder of the recording.



Until this post of yours, I didn't know anything about foobar2000, VST and MEqualizer. It took me some time to figure it out; eventually it's all installed.
I've been playing with it, and I can see how MEqualizer can help, but ... since I have zero understanding of all those settings, I am not getting anywhere.
It would be great if you, or someone else, would be able to advise with what settings I should play, or even better - to what values to change in order to get the best possible voice quality (and I do understand that anyway it will be limited since the original recording apparantly was lacking of quality environment) ...

Thank you ...

Help improving voice audio recordings

Reply #10
bad.mp3 sounds like it was transcoded from better.mp3.

I don't think better.mp3 was created by improving bad.mp3 - you can't remove that kind of distortion.

Cheers,
David.

 

Help improving voice audio recordings

Reply #11
bad.mp3 sounds like it was transcoded from better.mp3.

I don't think better.mp3 was created by improving bad.mp3 - you can't remove that kind of distortion.

Cheers,
David.



To be honest, I also thought about this possibility.
It is possible, though strange since their original version (not the one I cut and put the linked for) are the same size and same encoding. So why would someone do that, just add distortion ...?