I Found a free CFX soundfont and it works great on my PSR S970

[rodrigo.b]

Hi guys, here are a few demos of the free CFX sound https://soundcloud.com/rodrigo-bamonde-vizcarra/free-cfx-sound/s-IDdFGN2SdKP?si=34516db71e02407fb7302de2672cd5c4&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing I hope you like it. Make a comment here if you like it and I will share the sound.

[Amwilburn]

That sounds real! and really great playing, as well!

I'm unfamiliar with soundfont; is it like a vst, or a sample?

Mark

[rodrigo.b]

That sounds real! and really great playing, as well!

I'm unfamiliar with soundfont; is it like a vst, or a sample?

Mark

Thanks for comment. I found it on a YouTube video, but I have no idea if he sampled it from
A VST. I will make the PPF file so everyone can use it and I will post the link soon.

[rodrigo.b]

Here is the link for the PPF file, I'm also including a free Alto Sax that I found https://drive.google.com/file/d/1qHllVWxa3pxE5nFab4Kpi3hITGg33IX3/view?usp=sharing The original SF2 is from a YT Channel called "Lyzen's Piano"

[Keyboard Master]

Wow This is Amazing.

[rodrigo.b]

I found a another free piano SoundFont but this time it is a Yamaha C7 Soundfont. I imported the SF2 file in to YEM and tested it on my PSR S970. Here is the result: https://soundcloud.com/rodrigo-bamonde-vizcarra/free-c7-sf2-test-on-s970/s-oi6L4qzIrZT?si=98272bb811b6454687b8209fb09209c8&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing

[Amwilburn]

That one also sounds really nice; how do you import soundfont into pff? Just YEM? Another nice find!
How large would the ppf for that be? My poor s970 only has 509mb of room

(yes I know, time to upgrade to Genos. Wife said not until I sell off all my other boards )
Mark

[rodrigo.b]

That one also sounds really nice; how do you import soundfont into pff? Just YEM? Another nice find!
How large would the ppf for that be? My poor s970 only has 509mb of room

(yes I know, time to upgrade to Genos. Wife said not until I sell off all my other boards )
Mark

My 970 has 509 mb of memory too  . Yes I imported the soundfont with YEM. The C7 sound is 359 mb

[BogdanH]

First, thank you rodrigo for sharing your finding!

I installed Lyzen's CFX piano voice on SX700 and my first impression is, it does sound good/realistic -in short: different than any of preset pianos. So I decided to download complete Lyzen's collection (all kind of pianos: Yamaha, Bösendorfer, Kawai, etc.).
The problem is (as some of you mentioned), these voices are quite big in size and after installing, there's not much space left for additional voices.

Anyway, I was wondering what makes a single voice so big and so I extracted wav samples from CFX SF2 file. My impression is, the creator was quite "generous" with samples:
As it seems, there are two samples (for L & R channel) for each key on keybed. All samples have sampling rate of 64000Hz and all samples have duration of about 11-12 seconds.

Now.. if samples would have sampling rate of 44100Hz (CD quality), then that alone would make voice ~45% smaller.
The second thing is... I understand that low keys need long duration, because they have long sustain. However, keys on right side have much shorter sustain and so duration of about 6-7 seconds would be more than enough (as it is now, the last 5-6s contain silence anyway). Why that matters? Because every second increases size for about 100kB and if every wav sample would only have length that's actually needed, that would reduce total voice size additional by about 40% (my guess).

I though I should share my finding to those who are interested.
Bogdan

[DerekA]

I'm no expert, but I would have thought sampling multiple velocity layers every few keys was a better option than sampling every key.

[BogdanH]

@DerekA
I'm not an expert either, but yes, that's how it's usually done (to reduce the size and time required to record samples). Of course having separate sample for each key is way more accurate (is actually a copy of real piano). On the other hand, having a sample only every few keys does the job good enough, I think.
I will check other Lyzen's piano voices and then I'll see if it's worth trying to "optimize" some of these voices -that is, to reduce size substantially without (much) loss of quality.

Bogdan

[Amwilburn]

@DerekA
I'm not an expert either, but yes, that's how it's usually done (to reduce the size and time required to record samples). Of course having separate sample for each key is way more accurate (is actually a copy of real piano). On the other hand, having a sample only every few keys does the job good enough, I think.
I will check other Lyzen's piano voices and then I'll see if it's worth trying to "optimize" some of these voices -that is, to reduce size substantially without (much) loss of quality.

Bogdan

that would be fantastic! Yes, 12 second samples for the top 2 octaves is a complete waste, as is recording every single note (digitial pianos used to record only every 3rd note, and except for some odd chord harmonics on certain chords when sustaining, it was impossible to tell otherwise). Frankly I'd try optimizing with every 4th (next factor of 12 would be 6)

Mark

[BogdanH]

Ok, I made few piano voices from Lyzen's collection. I decided to use 4 samples per octave (means, one sample covers 3 notes), which then covers C0 to C6 octave.  After resampling to 44.1kHz, I shortened length of samples accordingly: lowest octave (C0) 12-14sec and all the way to highest octave (C6) 1.5sec. That way I ended with about 32MB per each voice (Yamaha CFX, Steinway D274, Bösendorfer 290, Fazioli F308 and Kawai EX-S). Btw. I think voice would sound good even with only 3 samples/octave, but I haven't tried that yet.

I didn't modify samples (no EQ, no Compression, etc.), because I wanted to hear samples as they are. As for voice settings in YEM, I left most of them at default -I only changed ADSR curve, so voice responds more realistic when playing.

Obviously I never heard these high class pianos in real life, but I do remember how a "real piano" sounds. Now, how do these voices sound to me?...
In my ears, much better than any piano voice on my keyboard. Now that I can hear how piano can sound on my keyboard, all built-in piano voices sound to me like coming from a toy: flat and dull.
Keep in mind, that this is my personal opinion!

Just thought that I share my experience and maybe encourage others to try making custom voices.

Bogdan

[Amwilburn]

Sweet, Bogdan! I know itunes can remix almost any audio format into 48khz or 44.1wav (I think Audacity can as well)... I wonder if 320kpbs mp3 files would also work as a soundfont? Then they could take up ~1/4 of the size!
What I don't know is how to take those reduced files and replace the original soundfont wavs, and them make a ppf from those? Could you elabourate?
Also would love to try the multiple other pianos from Lyzen, especially if you were able to get them down to 32mb each in *wav*!

thanks,

Mark

[BogdanH]

hi Mark,
First a short explanation for those who might peek into this thread... When we load any kind of audio file (mp3, flac, etc) into audio editor, it is first automatically converted to PCM audio data (except wav, which already contains PCM data). And the same happens when we playback any audio file. The reason for that is, because hardware (DAC) can only convert PCM data into analog audio. In that sense, mp3 is similar to zip file (except mp3 compression is much more effective for audio data).
Besides wav, Soundfont2 (SF2) editors usually only recognize lossless compressed audio files (i.e. flac). And it happens again: as soon flac file is imported into soundfont editor, it's converted to PCM. Means, even if it would read mp3 file, we wouldn't gain in file size. Ok, theoretically audio sample could be stored as mp3 inside soundfile (to make sf2 file smaller) -but there's a reason why that's not the case...
A voice (in YEM) is actually the same as soundfont and as we know, YEM only accepts wav samples. Yes, it could also accept mp3 files, but they would need to be converted to PCM data anyway.
Question remains: why aren't samples stored as mp3 compressed data inside voice (or inside soundfont)? The reason (in my opinion) is not because of sound quality.. it's because of speed. Because every time we would hit some key, sample would need to be decompressed (to PCM) first -which requires time. Now imagine playing 10 fingers, plus style is running... it would be impossible to keep the timing. PCM data doesn't need to be decoded/decompressed, because it's a stream of data that is directly sent to hardware and happens instantly.
The only way to reduce voice size is, to "optimize" wav samples:
1. To reduce number of samples per octave (I need to experiment more with this)
2. To reduce length of samples as much as possible
3. To reduce sample rate to 44100Hz (YEM doesn't accept lower than that).

Mark, I've sent you a mail few days ago and because you didn't respond I though you aren't interested.. I guess mail got lost somewhere at North pole  Email me, so I can send you the pack.

Bogdan

[rodrigo.b]

hi Mark,
First a short explanation for those who might peek into this thread... When we load any kind of audio file (mp3, flac, etc) into audio editor, it is first automatically converted to PCM audio data (except wav, which already contains PCM data). And the same happens when we playback any audio file. The reason for that is, because hardware (DAC) can only convert PCM data into analog audio. In that sense, mp3 is similar to zip file (except mp3 compression is much more effective for audio data).
Besides wav, Soundfont2 (SF2) editors usually only recognize lossless compressed audio files (i.e. flac). And it happens again: as soon flac file is imported into soundfont editor, it's converted to PCM. Means, even if it would read mp3 file, we wouldn't gain in file size. Ok, theoretically audio sample could be stored as mp3 inside soundfile (to make sf2 file smaller) -but there's a reason why that's not the case...
A voice (in YEM) is actually the same as soundfont and as we know, YEM only accepts wav samples. Yes, it could also accept mp3 files, but they would need to be converted to PCM data anyway.
Question remains: why aren't samples stored as mp3 compressed data inside voice (or inside soundfont)? The reason (in my opinion) is not because of sound quality.. it's because of speed. Because every time we would hit some key, sample would need to be decompressed (to PCM) first -which requires time. Now imagine playing 10 fingers, plus style is running... it would be impossible to keep the timing. PCM data doesn't need to be decoded/decompressed, because it's a stream of data that is directly sent to hardware and happens instantly.
The only way to reduce voice size is, to "optimize" wav samples:
1. To reduce number of samples per octave (I need to experiment more with this)
2. To reduce length of samples as much as possible
3. To reduce sample rate to 44100Hz (YEM doesn't accept lower than that).

Mark, I've sent you a mail few days ago and because you didn't respond I though you aren't interested.. I guess mail got lost somewhere at North pole  Email me, so I can send you the pack.

Bogdan

Please send me the pack too 🙂

[BogdanH]

hi rodrigo,
You started with this and so I would be happy to send it to you. But from what I can see, you didn't activated your eMail in this forum (files can not be attached to PM). You need to activate eMail first, so you can contact me (or give me your eMail in PM).

General disclaimer:
This is my personal experiment with voices and is meant for learning purposes.

For those who might try making custom voices from Lyzen's soundfonts pack:
Although voices can be used for playing just fine, later I realized that some wav samples contain "sound error". I corrected that where error was obviously audible, but I'm sure I missed some of them. What I'm saying is, it can be very time consuming to make "perfect" voice from Lyzen's samples. But it makes fun playing voices created by yourself 

Bogdan

[Amwilburn]

hi Mark,
First a short explanation for those who might peek into this thread... When we load any kind of audio file (mp3, flac, etc) into audio editor, it is first automatically converted to PCM audio data (except wav, which already contains PCM data). And the same happens when we playback any audio file. The reason for that is, because hardware (DAC) can only convert PCM data into analog audio. In that sense, mp3 is similar to zip file (except mp3 compression is much more effective for audio data).
*snip*

Mark, I've sent you a mail few days ago and because you didn't respond I though you aren't interested.. I guess mail got lost somewhere at North pole  Email me, so I can send you the pack.

Bogdan

*Smacks forehead*. Of course, pulse code modulation is still a linear lossless wav equivalent! D'uh, I'm getting old, lol.

I checked my emails, nothing... what was the title of the email? thanks! I'll see if I can find it; I checked Bogdan, PSRTutorial, soundfont and piano emails, nada.

Thanks for both!!
Mark

[BogdanH]

hi Mark,
I sent you email by clicking on email icon on your profile. I assume forum engine should send the message to your actual email and once you have it, you can see my email address to contact me. Is your email in this forum still valid? Just asking 
Or send me PM with your email address.

Btw. I know you have Genos (which isn't listed in your profile), but do you still have PSR-S970 (which is listed)?

Bogdan

[Amwilburn]

i play with Genos at work, but I own what I listed, incl PSRs970, yup.

ok pm-ing  you my personal email; yes clicking on the forum link should take you to the same email.. weird!

Mark

[BogdanH]

Will send you tomorrow.. it's 10PM here right now 

Bogdan

[Amwilburn]

Thanks Bodgan, got the pack!

I'll try it on the work Genos first,
Mark

[BogdanH]

Great.. just keep in mind that voices are in no way "customized/optimized" -they sound as it came into microphone. I'm curious about your opinion, of course.

Bogdan

[Amwilburn]

FInally got it to download. work hd was full (it's shared) so I had to delete some sample packs to be able to download yours!
A tiny 64mb, awesome! Until I unzipped it... holy that's a good compression ratio!

Hmm.. Couldn't test it. An error occurred and data could not be imported.


ACK! No wonder. I just realized I'm still on YEM2.81; you used 2.9 to create, correct?

lol!

Mark

[BogdanH]

Yes, I'm using YEM2.90
WinRAR can compress wav (which is 95% of voice pack) extremely well. Here's a comparison of compression (for archiving purposes), where I used 8min24sec 16bit/44100kHz audio track (Dire Straits-Why worry):
wav =86.736KB, rar =50.475KB, flac=37.082KB
-that's lossless compression* (means, no loss of quality). Btw. mp3 at 320kbps has 19.670KB (which is lossy compression).

*-wav is not compressed.

Hurry Mark, I'm curious about your opinion 
Bogdan

[SciNote]

Very interesting discussion.  But wouldn't over 8 minutes of WAV produce a file on the order of about 80 MB, not KB?

[overover]

Very interesting discussion.  But wouldn't over 8 minutes of WAV produce a file on the order of about 80 MB, not KB?

Hi Bob,

Bogdan's information is in KB (1024 KB = 1 MB), but to group the numbers, a comma would have to be used in English instead of the decimal point (i.e. exactly the opposite of e.g. in German).

For example, "wav =86.736KB" in Bogdan's post actually means 86736KB (= 86,736KB), which is about 84.7MB.)


Best regards,
Chris

[BogdanH]

Chris, thank you for explaining!
Yes, here we use a "point" as thousand separator and comma as decimal separator. We even say "decimal comma" (in my language) 

Bogdan

[SciNote]

Ah, yes. I forgot about how the functions of the comma and decimal point are switched throughout the world!

[Amwilburn]

Yes, I'm using YEM2.90
WinRAR can compress wav (which is 95% of voice pack) extremely well. Here's a comparison of compression (for archiving purposes), where I used 8min24sec 16bit/44100kHz audio track (Dire Straits-Why worry):
wav =86.736KB, rar =50.475KB, flac=37.082KB
-that's lossless compression* (means, no loss of quality). Btw. mp3 at 320kbps has 19.670KB (which is lossy compression).

*-wav is not compressed.

Hurry Mark, I'm curious about your opinion 
Bogdan

Updated to YEM 2.90.... still gives me an error occurred, could not import?

D'oh

Mark