Bit perfect editing

I am a sleep talker and I am recording what I say while I sleep. That means that I have dozens of huge audio files that are mostly empty. These files weigh about 200 MB each (8 hours long). I want to edit the files to remove all the silences and the noises, and leave only the sleep talk. I want to do this in bit perfect mode to prevent losing quality in the process. I’d like to know if I can do this with Audacity, and if not, what other software exists that allows me to do this.

I have already tried WaveShop, and it works well, but its functionality is very limited and is not very user friendly.

Note: I have just realized something quite obvious: There are recording apps that skip silences automatically. I have just downloaded one of these, and if it works well I will be using it from now on. But I still want to process the huge files that I have recorded so far.

200 MB each (8 hours long).

If the shows are really 200MB for eight hours, then it’s already too late for bit-perfect editing. The shows have been saved in a compressed format and most compressors add distortion. It gets worse. Because the shows are compressed, chances are pretty good they’re in MP3 and Audacity always causes additional damage when you edit MP3s. It’s burned in and you can’t stop it.

You should find out what format the shows are in (if you don’t already know). Media Info is a good choice for this.

Then search for an editor that works intimately with that format. If it’s MP3, people have found MP3Splt to be valuable.

Koz

Editing an Audacity project is always “bit perfect”. What may not bit perfect is converting between Audacity projects and audio file formats.

Importing or exporting 32 bit float WAV format is completely lossless but the file size is huge.
Importing / Exporting 16 bit FLAC format is extremely high quality, though not technically “bit perfect”. A 16 bit FLAC file is likely to be about 1/4 of the size of a 32 bit float WAV format file.

MP3, Ogg, WMA and other “compressed” formats are able to give “good”, or even “excellent” sound quality (depending on the compression settings) but are (by definition) not bit perfect.

Yeah, the audio is compressed and distorted already, of course. I just want to avoid adding more fiddling and distortion with my editing.

This sounds very promising. The format is MP4 embedded in 3GP. Do you know of a bit perfect MP4 editor?

Search Google for “3gp splitter”.

Take care when downloading and installing - it is common for “free” downloads to either not really be free, or to be bundled with useless or even dangerous other software. You should only download programs that you trust.

Some advanced compression formats can’t be “lossless edited.” In addition to the usual tricks of Fourier stripping and bit restrictions, they depend on re-arranging the show’s data for even smaller files. Cut the show’s data without decompression/recompression and it falls apart.

Compression formats make terrible production material, a cousin to “never edit in MP3.”

Koz

I have searched Google, and there seems to be several programs by the name 3GP splitter. Can you be more specific? Maybe give a link?

Sorry, no. I don’t use GP3 format and have not used any of those programs, but if you want “lossless” editing of a file that is in a “lossy” format, you will need a program that can perform edits without decoding / re-encoding the file. Audacity can’t do that. Audacity always works in high quality uncompressed audio, which means that lossy compressed formats must be decoded first, then (if the compressed format is required) re-encoded on export.

The idea of “bit perfect editing” is somewhat undermined by the fact that you have already subjected the audio to losses due to the encoding.

Thank you for your help. After investigating a bit (but only a little bit), it looks like the best option is to use Audacity to find the pieces that I want to keep and then Waveshop to do the actual chopping. On a related note, I’d like to know if there is some way to locate automatically the parts of a recording that include voice in a huge file. Once again, I’m talking about files that are several hours long. They include a lot of silence (no problems here), a lot of noise (this is the problem) and some voice, which is tedious to isolate manually.

Ideas?

Audacity does not have voice recognition.

If the voice is substantially louder than the background noise, then you may be able to add labels where the voice occurs using Sound Finder: http://manual.audacityteam.org/o/man/silence_finder_and_sound_finder.html#sound
However, I don’t know how well Sound Finder will cope with a recording that is “hours long”. It will certainly be slow, and if your computer does not have enough free ram it may crash, so you may be better to process the file in sections.

The other thing that I would suggest is to just visually inspect the track. If you zoom in to an appropriate scale you may be able to identify the where the voice occurs visually.
Zooming: http://manual.audacityteam.org/o/man/zooming.html

First off, I forgot to show appreciation for your answers a few years ago, so thank you for your help.

Second, I know that this is a very old thread, but I have a little question about it that doesn’t really require a new thread:

Is it also lossless to import and export 16 bits WAV format? I’m currently using a recorder that saves the audio files in 16 bits WAV by default, so it would be very handy if it is.

There are recording apps that skip silences automatically. I have just downloaded one of these, and if it works well I will be using it from now on.

How would you know?

Let me read this back to you. If the software fails to record me reliably at night, I’ll know not to use it any more.

Silent Sense programs all have sensitivity settings. Nobody’s bedroom is completely silent, so the software has to decide which sound is normal background and which are performance. Throw normal microphone noise in there and chance of reliable recording is very poor.

You could double record. One recorder has the software on it and the other works the way yours does now. Play them both and if the new software performs as you want, start using it.

Otherwise, you’ll never know.

Koz

Ah, don’t worry, I dealt with that problem long ago. You’re correct, of course. If I set the sound threshold too low, the recorder detects and records even the slightest noises. If I set it too high, the recorder misses some of my sleep talking. But it hasn’t been a big deal for me because I don’t really need to record 100% of my sleep talking. 80% or so is just fine. So I set the sensitivity to a medium spot and it works pretty well for me.

Still, even using the silence skipping feature I end up with audio files that last between one and two hours because, as you said, my bedroom is not completely silent, so the recorder still records plenty of noises. Only about 5% of the recording is useful, the rest is just noise. So I edit the file manually to select the 5% that is useful and remove the rest. I want to do that editing in a lossless (or bit perfect) way. I can do that with WaveShop, but it’s a pain because it’s not remotely as user friendly as Audacity, so I’ve ended up doing it with Audacity. However, I’m concerned that Audacity might be doing that in a lossy way, introducing recompression and distortion. Hence my question: Does Audacity import and export the 16 bits WAV format in a lossless or a lossy way?

That question is not as straightforward as it might appear.

The short answer is that it is lossless, but with caveats:

Audacity works internally in 32-bit float format. When converting from a higher bit format to a lower bit format, “dither” should normally be applied so as to avoid “quantize errors” (rounding errors).

When importing a 16 bit file, by default Audacity will convert the audio data to 32-bit float, which is a lossless conversion (essentially just padding the data with some zeros).

When exporting to a 16-bit file, Audacity converts the 32-bit float data to 16-bit integer, and by default Audacity applies “dither”. Dither may be turned off in Audacity Preferences if required, but normally it should be enabled for best audio quality.

More information about dither: Missing features - Audacity Support

Just wanted to say thank you for your answer, steve. The way I understand it, the noise introduced by the dithering is so small that I can consider that this is lossless edition for my purposes (I don’t need very high quality).

I’ve done night noise tests where I took what I knew was “clean” background noise and did a simple Noise Reduction. The overall hash dipped to the point where I could find the lumps and bumps in the blue waves enough to note the times and go in later and inspect them. No gating or voice recognition needed.

My recorder did it in perfect quality, uncompressed wav (like yours?) and the recorder knows to start a new file when it hits 2GB.

If you have a noisy neighborhood, none of these simple tricks are going to work.

Koz

Thank you for the feedback. My neighbourhood is mostly quite, so that’s not a problem. But the noises are very irregular. I snore, my cat meaows, I turn around in bed and the bed makes noises… I can’t remove that kind of noises with standard noise reduction. Anyway, these days I don’t make these recordings so often, so I’m fine with a bit of manual editing.

I snore, my cat meaows, I turn around in bed and the bed makes noises

The simple noise reduction trick makes all those stick up so you can do rapid logging. This is a bookkeeping chore.

Snore.
Snore.
Cat.
Trip to the loo.
Masters Dissertation from 15 years ago.



3:09 Messing with the pillow.
3:17 Turning Over.
3:18 Yawning.

Koz

Once you get to a comfortable zoom range, Shift+ScrollWheel will let you slide left and right. If you’re gentle, you can catch all those little lumps and bumps as they go by. As you get better at it, you will start to recognize patterns. Too short. Too Short. This one looks Promising.

Snoring tends to follow REM Sleep cycles, so those can be 90 minute predictable. If they’re not predictable, you should be tested for Apnea.

Koz