Oops - wrong sample rate - help ?

[Lizardpoint]

Morning team,
I just got some vocals back from a singer on Fiverr. The vocals are great but out of tune/with my track back at this end.
I think they imported my 48k track into a 44.1k session and then sang to that. The result is although they sound great, they are unusable. Is there any clever way I can save this or will they need to be redone ?
Cheers
Dave

[The Elf]

Been here. Done this.

You need to re-tune the audio they've returned to you. Problem is - did they work to a time-stretched version of the audio you sent?

Get them to send an export of the audio they worked to, so you can hear it as they heard it. Import it into your project and see if it matches in time. Then you can see if stretching needs to be done.

If it matches in time, but not in pitch then stretching will be required. If it doesn't match in time then it may simply be a case of re-pitching the audio without stretching.

And then there's the maths involved in calculating the new pitch. I have to re-learn this every time I do it, but essentially it involves working out out one of those sample rates as percentage of the other and using this to re-pitch the audio you want to keep. Hopefully someone will be along here with the formula!

I'd advise stretching both the added audio and the backing they send at the same time. If the re-pitched/stretched backing matches your project then so should the added audio.

[Wonks]

Is it worth just trying a sample rate converter on one file so see if that fixes the issue before trying to pitch shift and stretch?

[Sam Spoons]

I'd just add, how hard will it be to get it redone? Unless it was a brilliant, unrepeatable performance redoing it will always give the best results.

If it's from fivver they should probably redo it for free as the error was at their end but they may not be happy with that?

[The Elf]

Is it worth just trying a sample rate converter on one file so see if that fixes the issue before trying to pitch shift and stretch?

That's essentially what I'm suggesting. Re-pitching without stretch is really just sample rate conversion. You really need the backing they worked to to be sure of what you're hearing, though.

[Hugh Robjohns]

Sample rate conversion doesn't change pitch or duration. It resamples at a new rate the waveform reconstructed from the first sample rate.

In the situation above the original backing track file has been replayed at the wrong rate, hence the pitch has changed. Just like varispeeding a tape machine.

Not sure how they managed that. Most DAWs will automatically SRC a file if its at a different sample rate to the project when its imported.

48 to 44.1k (or vice versa) is about an 8% varispeed change.

Ideally, I'd get the vocals re-recorded properly. You can varispeed the vocal to fit, but the formats will move and it won't sound like the original voice anymore. It may be unnoticeable to most, but I bet the singer will know!

[The Elf]

Sample rate conversion doesn't change pitch or duration. It resamples at a new rate the waveform reconstructed from the first sample rate.

I shouldn't have used the word 'conversion' above. What I'm suggesting is changing the playback rate.

Yep, it's going to alter the tone of a vocal. Maybe not so important for other instrumentation.

[Hugh Robjohns]

[ken long]

Not sure how they managed that. Most DAWs will automatically SRC a file if its at a different sample rate to the project when its imported.

Probably a dialogue upon import asking if they wanted to change the session sample rate to match the imported file and they chose "no". Some DAWs don't subsequently and automatically convert the audio.

[Tomás Mulcahy]

What a mess. If they're a singer who is worth using they will be able to sing it the same again.

But there is a workaround. I sometimes have to deal with issues like this with collaborators. It's different for a paid service like you used, so I am not suggesting this as the best solution!

Do what Elf suggested, get the actual file/ DAW project and media they sang over. Drop that file, their vocal file , and the correct backing track into Melodyne Studio. It will line them all up for you. Manually correct pitch if necessary. Now export the vocal alone at the SR you need.

[merlyn]

To work out the change in pitch in cents the formula is

1200×log2(f2/f1)

In this case 1200×log2(48/44.1) = ~146.7 cents

Doing the calculation is complicated by the fact that calculators usually don't have a 'log to the base 2' button. So we use the property loga(x) = logb(x)÷logb(a). Calculators have a natural log button -- ln and a 'log to the base 10' button. So the calculation above in full is :

1200×ln(48/44.1)÷ln(2) = ~146.7 cents

That is nearly a semitone and a half sharp.

[Hugh Robjohns]

I love it when people show their working...

[The Elf]

So, for those of us who struggle with these things...

What calculation do I need to apply in percentage to convert:
44.1 to 48
48 to 44.1

And what calculation do I need to do to apply in +/- pitch values (i.e. 0.0 being no change) to convert:
44.1 to 48
48 to 44.1

I think many of us would appreciate it if you could give us the simple version that non-scientific calculators can handle.

As I mentioned, I've had to re-learn this every time I've had to do it. On the last album I had to re-pitch a backing track that had been shifted by an incorrect tape speed back in the 70s... re-pitch this to enable a whistle/pipes player to add his parts (since he couldn't re-tune his instruments)... and bring the player's audio back and re-pitch it to the original out-of tune tape version AND a corrected version to which I'd added my own parts.

Needless to say the maths for all this still has me waking up screaming in the night!

[merlyn]

The percentage is easy enough :

48÷44.1×100 = ~108.84 that's an increase of ~8.84%

44.1÷48×100 = 91.875 that's a decrease of 8.125%

The relationship between frequency and pitch is logarithmic, so you need logs to calculate a pitch from frequencies. Tough luck.

[The Elf]

The relationship between frequency and pitch is logarithmic, so you need logs to calculate a pitch from frequencies. Tough luck.

Dear lord! That's me out then!

[The Elf]

[Wonks]

User error

[The Elf]

[merlyn]

Eh ... I don't know where you went wrong there but it's 146.7 cents, and -146.7 cents.

That may help get your head around logs. (Or not). With the percentages the increase and decrease are different , but with logs, because we're dealing with ratios, only the sign is reversed.

[The Elf]

I've saved those end numbers - the calculation I'll just smile at nicely and move on...

Hopefully they will help the OP.

[merlyn]

It's really not difficult. log2 measures how many octaves there are in a frequency ratio.

1 2 4 8

Those numbers are going up in octaves. log2(1/1) is 0. There are no octaves in that. log2 is the answer to the question "what power do I have to raise 2 to to get the number I want?" log2(2/1) is 1. There is one octave in that ratio. log2(4/1) is 2. There are two octaves in that frequency ratio. log2(8/1) is 3. There are three octaves in that ratio. Then multiply by 1200 to get the number of cents, as there are 1200 cents in an octave.

But sure, 146.7 cents is the answer to sampling rate mess-ups.

[Wonks]

Eh ... I don't know where you went wrong there but it's 146.7 cents, and -146.7 cents.

I noticed I'd multiplied by ln(2) rather than divided, hence the wrong value. I have a habit of clicking the wrong function when using the Windows calculator.

[RichardT]

If it's certain the singer's software has interpreted a 48kHz audio file as 44.1kHz, you just need to do the opposite process on their vocal recording.

As people have said, this is not a normal thing to happen when importing audio into a DAW, but I guess that some other software might have changed the sample rate of their audio device behind the scenes and screwed things up in the DAW.

What you could do on the vocal recording is use something like Izotope RX to update the sample rate metadata on the file from 44.1 to 48 without actually resampling anything.

Then import to the original project.

The vocals will appear at the right pitch. However, they will not sound like the singer sang them at that pitch. It's probably worth a go though.

If you don't have RX or an equivalent, drop me a pm with a link and I'll do it for you.

[Tomás Mulcahy]

It's really not difficult.

It is if someone has dyscalculia. Like me It takes about 10 times longer to figure out, and I need to re-learn it every time I need to use it again.

[merlyn]

Multiplication doesn't seem to cause so many problems. Multiplication is repeated addition. 4×3 means 4+4+4. The '×' is shorthand for repeated addition. 4×3 means 'add four to itself three times'. The inverse operation is division. We can get back to where we started with division. 4×3 = 12. 12÷3 = 4.

Powers of a number are repeated multiplication. 4^3, spoken as 'four cubed', means 4*4*4. The '^' is shorthand for repeated multiplication. 4^3 means 'multiply four by itself three times'. The inverse operation is 'log to the base of'. We can get back to where we started by taking a log. 4^3 = 64. log4(64) = 3.

That's how I understand it, and powers and logs are relevant to audio because of dBs and pitch.