6 ms·

> You can now change the duration of your audio clips without affecting their pitch non-destructively. Audacity 3.4 uses a new time stretching algorithm specifi

by Aachen 3y ago

> You can now change the duration of your audio clips without affecting their pitch non-destructively. Audacity 3.4 uses a new time stretching algorithm specifically made for music, which outperforms many of the commercially available options. Way to bury the lede! What's this magic algorithm being spoken of, how does it work so well? > A more detailed overview of these changes can be found in our <a>changelog</a>. The changelog is even more terse, saying only which key combo activates it :( A way bigger deal to me is mentioned at the bottom of the "other changes" in said changelog: Opus support! Edit: wait, under "libraries" it says > lib-time-and-pitch implements a time stretching algorithm originating in Staffpad. It currently is one of the highest-quality time stretching algorithms for music on the market. Staffpad links to https://staffpad.net https://staffpad.net and seems to be some music app. The library name links to a Microsoft repository (https://github.com/audacity/audacity/tree/e4bc052201eb0e6e22956cb6426e88661713c6d6/libraries/lib-time-and-pitch https://github.com/audacity/audacity/tree/e4bc052201eb0e6e22...) with no readme. The source code is documentation enough :)

dmix 3y ago
> a new time stretching algorithm specifically made for music, which outperforms many of the commercially available options. I wonder how it compares to Ableton Live, warping was always a big part of Abelton.
swyx 3y ago
it would be nice to at least know in what ways it "outperforms many of the commercially available options.". is this asking too much?
ksherlock 3y ago
This is the PR I believe Time stretching 2 of 6 make audio track stretching effective https://github.com/audacity/audacity/pull/5041 https://github.com/audacity/audacity/pull/5041
beepbooptheory 3y ago
I just feel like, for just stretching stuff and staying in pitch, any off-the-shelf FFT-based method has near perfect results within a meaningful range anyway. It has felt like a solved problem for a while? What does the "high-quality" mean here?
- colanderman 3y ago
  Most stretching algorithms I've heard have noticeable "ghosting" of transients, and tend to muddle synthesizer tones (due to discarding phase information). I am very curious to test this one.
Blackthorn 3y ago
Is there a comparison of different algorithms and how they work anywhere? I only know of two, and I'd really like to understand them better. The two I know of are Paulstretch (which is really only for super-duper-long stretching) and Lent 1989 ("An Efficient Method for Pitch Shifting Digital Sampled Sounds"), which can be thought of as a form of granular synthesis but isn't really what most people think of when they hear that term.
willis936 3y ago
Is this different than paul stretching with infinite duration?
- Slow_Hand 3y ago
  Yes. The Audacity algorithm is a granular time-stretch, whereas PaulStretch is an FFT Transform for time-stretching. Depending on your needs, you'd want to favor one over the other. Granular stretches are far less CPU intensive and have significantly lower latency than an FFT Transform. The granular algorithm will likely have better transient fidelity at small time-stretch intervals (between 0.5x - 2x speeds), whereas FFTs tend to smear the transient information. Where FFT transforms really excel is in maintaining or transforming formants and harmonic information. Especially at extreme pitch or time settings. Granular algorithms can only stretch so far before you hear the artifacts. FFTs are far more graceful for dramatic transforms.
TheAceOfHearts 3y ago
Tangential to this, I've noticed significant differences in audio quality when listening to audiobooks at 2x speed on VLC vs MPV. In particular MPV doesn't seem as clipped. A bit of an open ended question, but is there anything more I could do to process the audiobook to make it sound even better at 2x?
- ddejohn 3y ago
  > A bit of an open ended question, but is there anything more I could do to process the audiobook to make it sound even better at 2x? Sounds like you could download Audacity 3.4 and make a 2x version of your audiobook files using their new time stretching algorithm.
- fbdab103 3y ago
  2x! What kind of books are you reading? I have gone as far as 1.4-1.5x, but that feels about the best I can do without having to be laser focused on just listening (ie not doing chores or other minor activities as is my habit).
Slow_Hand 3y ago
I was curious myself, so I downloaded Audacity and tried. It's a granular time-stretch algorithm. By the sound of the artifacts at extreme settings it's akin to the 'Complex Pro' algorithm in Ableton Live (if you have any reference to that) and seems to be equally effective at a wide variety of sources (percussion, vocals, full records). Is it better than most commercial offerings? Hard to say after a brief test, but it's not bad! I suspect it's plenty good for the needs of Audacity's audience, who are unlikely to be demanding audio pros. As an audio professional I would never use Audacity, but if you need a quick, free, and (relatively) simple option to time-stretch a file, then it should fit the bill.
- westcoast49 3y ago
  > better than most commercial offerings The audacity!
  - thowawa47827 3y ago
    <3
- vintermann 3y ago
  Do you really make enough from being an audio professional that price doesn't matter? Whether you do or not, there are a hell of a lot of hawkers of software and "hardware" alike, who stand ready to sell their overpriced "professional" tools to people who fail to make it in the music industry (which is going to be most of them, regardless of tools or even skills). A high sticker price is a way to prove your commitment - to oneself, at least. The end market is less likely to care.
  - hurryer 3y ago
    Good professional audio hardware is not overpriced. Sure, there are $15k microphones but you absolutely do not need those, a $1k one is top quality and not expensive if you use it regularly. You are confusing it with audiophile gear. Regarding Audacity, it's missing tons of features and it's UI is bad. Just like Gimp vs Photoshop. I did some audio production and Audacity is just not usable beyond the most basic operations, and even those are a bit of pain.
bdhcuidbebe 3y ago
> The library name links to a Microsoft repository. You link to the official audacity and chose to (incorrectly) describe it as ”a Microsoft repository” Why being dishonest?
- Aachen 3y ago
  If I say GitHub repository, does the sentence then make sense to you? I tend to forget that GH is Microsoft's similar to how people around here forget WhatsApp is Facebook or random brands are Nestlé, so I try to get in the habit of calling things by the parent brand to decrease the obfuscation Maybe I should simply have said git, though. Didn't think of that until now at the end of writing this reply. No dishonesty intended, quite the opposite in fact
  - bdhcuidbebe 3y ago
    ”a microsoft repository” implies that the audacity project is handled by microsoft, which false. If you must state ownership you could have done so with something like ”a microsoft-hosted repository”. Most of us have been around since before microsoft bought github.
jai_ 3y ago
Staffpad and Audacity are both software under the ownership (?) of Muse Group. Staffpad is recent acquisition which is why they are now able to share technology like this: https://mu.se/ https://mu.se/
heycosmo 3y ago
The time stretch algorithm is implemented in https://github.com/audacity/audacity/blob/master/libraries/lib-time-and-pitch/StaffPad/TimeAndPitch.cpp https://github.com/audacity/audacity/blob/master/libraries/l... particularly functions _time_stretch and _process_hop. It looks to me like a classic phase vocoder with vertical phase coherence (c.f. https://en.wikipedia.org/wiki/Phase_vocoder https://en.wikipedia.org/wiki/Phase_vocoder). The basic idea is this. For a time-stretch factor of, say, 2x, the frequency spectrum of the stretched output at 2 sec should be the same as the frequency spectrum of the unstretched input at 1 sec. The naive algorithm therefore takes a short section of signal at 1s, translates it to 2s and adds it to the result. Unfortunately, this method generates all sorts of unwanted artifacts. Imagine a pure sine wave. Now take 2 short sections of the wave from 2 random times, overlap them, and add them together. What happens? Well, it depends on the phase of each section. If the sections are out of phase, they cancel on the overlap; if in phase, they constructively interfere. The phase vocoder is all about overlapping and adding sections together so that the phases of all the different sine waves in the sections line up. Thus, in any phase vocoder algorithm, you will see code that searches for peaks in the spectrum (see _time_stretch code). Each peak is an assumed sine wave, and corresponding peaks in adjacent frames should have their phases match.
- QuantumGood 3y ago
  The old "Change Tempo" effect still works much better for voice.

by Aachen 3y ago

dmix 3y ago

> a new time stretching algorithm specifically made for music, which outperforms many of the commercially available options. I wonder how it compares to Ableton Live, warping was always a big part of Abelton.

swyx 3y ago

it would be nice to at least know in what ways it "outperforms many of the commercially available options.". is this asking too much?

ksherlock 3y ago

This is the PR I believe Time stretching 2 of 6 make audio track stretching effective https://github.com/audacity/audacity/pull/5041 https://github.com/audacity/audacity/pull/5041

beepbooptheory 3y ago

I just feel like, for just stretching stuff and staying in pitch, any off-the-shelf FFT-based method has near perfect results within a meaningful range anyway. It has felt like a solved problem for a while? What does the "high-quality" mean here?

colanderman 3y ago
Most stretching algorithms I've heard have noticeable "ghosting" of transients, and tend to muddle synthesizer tones (due to discarding phase information). I am very curious to test this one.

Blackthorn 3y ago

Is there a comparison of different algorithms and how they work anywhere? I only know of two, and I'd really like to understand them better. The two I know of are Paulstretch (which is really only for super-duper-long stretching) and Lent 1989 ("An Efficient Method for Pitch Shifting Digital Sampled Sounds"), which can be thought of as a form of granular synthesis but isn't really what most people think of when they hear that term.

willis936 3y ago

Is this different than paul stretching with infinite duration?

Slow_Hand 3y ago
Yes. The Audacity algorithm is a granular time-stretch, whereas PaulStretch is an FFT Transform for time-stretching. Depending on your needs, you'd want to favor one over the other. Granular stretches are far less CPU intensive and have significantly lower latency than an FFT Transform. The granular algorithm will likely have better transient fidelity at small time-stretch intervals (between 0.5x - 2x speeds), whereas FFTs tend to smear the transient information. Where FFT transforms really excel is in maintaining or transforming formants and harmonic information. Especially at extreme pitch or time settings. Granular algorithms can only stretch so far before you hear the artifacts. FFTs are far more graceful for dramatic transforms.

TheAceOfHearts 3y ago

Tangential to this, I've noticed significant differences in audio quality when listening to audiobooks at 2x speed on VLC vs MPV. In particular MPV doesn't seem as clipped. A bit of an open ended question, but is there anything more I could do to process the audiobook to make it sound even better at 2x?

ddejohn 3y ago
> A bit of an open ended question, but is there anything more I could do to process the audiobook to make it sound even better at 2x? Sounds like you could download Audacity 3.4 and make a 2x version of your audiobook files using their new time stretching algorithm.
fbdab103 3y ago
2x! What kind of books are you reading? I have gone as far as 1.4-1.5x, but that feels about the best I can do without having to be laser focused on just listening (ie not doing chores or other minor activities as is my habit).

Slow_Hand 3y ago

I was curious myself, so I downloaded Audacity and tried. It's a granular time-stretch algorithm. By the sound of the artifacts at extreme settings it's akin to the 'Complex Pro' algorithm in Ableton Live (if you have any reference to that) and seems to be equally effective at a wide variety of sources (percussion, vocals, full records). Is it better than most commercial offerings? Hard to say after a brief test, but it's not bad! I suspect it's plenty good for the needs of Audacity's audience, who are unlikely to be demanding audio pros. As an audio professional I would never use Audacity, but if you need a quick, free, and (relatively) simple option to time-stretch a file, then it should fit the bill.

westcoast49 3y ago
> better than most commercial offerings The audacity!
- thowawa47827 3y ago
  <3
vintermann 3y ago
Do you really make enough from being an audio professional that price doesn't matter? Whether you do or not, there are a hell of a lot of hawkers of software and "hardware" alike, who stand ready to sell their overpriced "professional" tools to people who fail to make it in the music industry (which is going to be most of them, regardless of tools or even skills). A high sticker price is a way to prove your commitment - to oneself, at least. The end market is less likely to care.
- hurryer 3y ago
  Good professional audio hardware is not overpriced. Sure, there are $15k microphones but you absolutely do not need those, a $1k one is top quality and not expensive if you use it regularly. You are confusing it with audiophile gear. Regarding Audacity, it's missing tons of features and it's UI is bad. Just like Gimp vs Photoshop. I did some audio production and Audacity is just not usable beyond the most basic operations, and even those are a bit of pain.

bdhcuidbebe 3y ago

> The library name links to a Microsoft repository. You link to the official audacity and chose to (incorrectly) describe it as ”a Microsoft repository” Why being dishonest?

Aachen 3y ago
If I say GitHub repository, does the sentence then make sense to you? I tend to forget that GH is Microsoft's similar to how people around here forget WhatsApp is Facebook or random brands are Nestlé, so I try to get in the habit of calling things by the parent brand to decrease the obfuscation Maybe I should simply have said git, though. Didn't think of that until now at the end of writing this reply. No dishonesty intended, quite the opposite in fact
- bdhcuidbebe 3y ago
  ”a microsoft repository” implies that the audacity project is handled by microsoft, which false. If you must state ownership you could have done so with something like ”a microsoft-hosted repository”. Most of us have been around since before microsoft bought github.

jai_ 3y ago

Staffpad and Audacity are both software under the ownership (?) of Muse Group. Staffpad is recent acquisition which is why they are now able to share technology like this: https://mu.se/ https://mu.se/

heycosmo 3y ago

The time stretch algorithm is implemented in https://github.com/audacity/audacity/blob/master/libraries/lib-time-and-pitch/StaffPad/TimeAndPitch.cpp https://github.com/audacity/audacity/blob/master/libraries/l... particularly functions _time_stretch and _process_hop. It looks to me like a classic phase vocoder with vertical phase coherence (c.f. https://en.wikipedia.org/wiki/Phase_vocoder https://en.wikipedia.org/wiki/Phase_vocoder). The basic idea is this. For a time-stretch factor of, say, 2x, the frequency spectrum of the stretched output at 2 sec should be the same as the frequency spectrum of the unstretched input at 1 sec. The naive algorithm therefore takes a short section of signal at 1s, translates it to 2s and adds it to the result. Unfortunately, this method generates all sorts of unwanted artifacts. Imagine a pure sine wave. Now take 2 short sections of the wave from 2 random times, overlap them, and add them together. What happens? Well, it depends on the phase of each section. If the sections are out of phase, they cancel on the overlap; if in phase, they constructively interfere. The phase vocoder is all about overlapping and adding sections together so that the phases of all the different sine waves in the sections line up. Thus, in any phase vocoder algorithm, you will see code that searches for peaks in the spectrum (see _time_stretch code). Each peak is an assumed sine wave, and corresponding peaks in adjacent frames should have their phases match.

QuantumGood 3y ago
The old "Change Tempo" effect still works much better for voice.