sfzen 2.2.1

Simple object-oriented SFZ parsing and manipulation.

sfzen

Intro

The SFZen package provides several scripts for SFZ inspection and maintenance, and a simple object-oriented SFZ parsing and manipulation library which you can use in your own scripts.

Credits

This project borrows a good deal from sfzlint by J. Isaac Stone. I made everything as object-oriented and pythonic as I could, and added a few methods which sfzlint didn't cover. Many thanks to "jisaacstone" for doing the ground work. Anyone who contributes to the Linux music community deserves recognition!

Command-line tools

There are several command-line scripts provided, including:

sfz-copy

Copies an SFZ to another location with multiple ways of handling samples.

Basic invocation:

sfz-copy sfzs/Foo-Piano.sfz projects/Bar/

This will copy the SFZ file "Foo-Piano.sfz" into the "projects/Bar/" folder. All the samples referenced in the original SFZ will be copied to a folder named "Foo-Piano-samples". The structure of the copied files will be:

─┬─ Foo-Piano.sfz
 ├─ Foo-Piano-samples/
 └─┬─ sample-1.wav
   ├─ sample-2.wav
   └─ ...

If you would like to make save space, as when you might be editing the target SFZ, but have no need to modify the original samples, there are several options which allow you to do that:

--symlink,  -s  Create symlinks in the target samples folder.
--hardlink, -l  Hardlink samples in the target samples folder.
--abspath,  -a  Point to the original samples - absolute path.
--relative, -r  Point to the original samples - relative path.

See "Sample modes when using "save_as" below.

The sfz-copy script also gives the option to "simplify" the SFZ during the copy.

"SFZ simplification" below

sfz-validate

Checks whether SFZ files are formatted correctly, and opcode values are of the correct type and have valid values.

To validate a single SFZ:

sfz-validate sfzs/Foo-Piano.sfz

This will print detailed errors, one per line:

sfzs/Foo-Piano.sfz:37 "lovel=0": Value [0] is below the minimum [1] for "lovel"
sfzs/Foo-Piano.sfz:39 "volume=10": Value [10.0] is above the maximum [6.0] for "volume"

If you would like a simple summary, use the "--summary" option:

    $ sfz-validate --summary sfzs/Foo-Piano.sfz
        1 MinError
        1 MaxError

To see exactly what values caused common errors to occur, use the "--summary" option twice:

$ sfz-validate -s -s --recurse sfzs/
   34 MaxError
	   2 fillfo_depth=1506
	   2 fillfo_depth=2528
	   [ .. ]
	8 MinError
	   2 fillfo_depth=-1302
	   1 fillfo_depth=-7200
	   [ .. ]

You may pass several filenames on the command line. To validate all the SFZs found inside a folder and all of its subfolders, use the "--recurse" option:

$ sfz-validate --summary --recurse sfzs/
   34 MaxError
	8 MinError

sfz-opcode-usage

Lists all of the opcodes used in one or many .sfz files.

Basic invocation (recursing to subfolders):

$ sfz-opcode-usage -r sfzs/
lokey
hikey
[ .. ]
amp_velcurve_1
amp_velcurve_127

sfz-samples

Prints paths to samples used in a given SFZ.

Basic invocation:

$ sfz-samples sfzs/VSCO/Harp.sfz
KSHarp_G3_mf.wav
KSHarp_D6_mf.wav
[ .. ]

Print absolute paths:

$ sfz-samples --abspath sfzs/VSCO/Harp.sfz
/mnt/data-drive/sfzs/VSCO/Harp/KSHarp_A4_mf.wav
/mnt/data-drive/sfzs/VSCO/Harp/KSHarp_F6_mf.wav
[ .. ]

sfz-opcode-info

Displays data type, allowable values, aliases and descriptions of opcodes.

$ sfz-opcode-info loopmode
loopmode
--------
Opcode name: loop_mode
Category:    Sample Playback
Version:     v1
Type:        str
Choices:
        no_loop         no looping will be performed. Sample will play straight from start to
                        end, or until note off, whatever reaches first.
        one_shot        sample will play from start to end, ignoring note off. This is
                        commonly used for drums. This mode is engaged automatically if the
                        count opcode is defined.
        loop_continuous once the player reaches sample loop point, the loop will play until
                        note expiration.
        loop_sustain    the player will play the loop while the note is held, by keeping it
                        depressed or by using the sustain pedal (CC64). The rest of the sample
                        will play after note release.
Default:     no_loop
Description:
   Allows playing samples with loops defined in the unlooped mode.

sfz-liquid-safe

Strips an SFZ of opcodes which liquidsfz does not support.

API

Basic Usage

from sfzen import SFZ

sfz = SFZ(filename)
... do stuff ...
sfz.write(sys.stdout)

Example

Let's say you have recorded a sample of each guitar string being plucked on your guitar. You have saved the sample files in a directory named "samples" beneath the current working directory. The sample files have the following names:

40-E2.wav
45-A2.wav
50-D3.wav
55-G3.wav
59-B3.wav
64-E4.wav

The number in the filename is the MIDI pitch of the string recorded.

The following script will create a very basic SFZ file from these samples:

from sys import stdout
from operator import itemgetter
from pathlib import Path
from sfzen import SFZ, Global, Region

sfz = SFZ()
global_header = sfz.append(Global())
global_header.loop_mode = 'one_shot'
tuples = [
	(int(sample_path.name[0:2]), sample_path)
	for sample_path in Path('samples').rglob('*.wav')
]
tuples.sort(key = itemgetter(0))
for index, tup in enumerate(tuples):
	pitch, path = tup
	region = global_header.append(Region())
	region.sample = path
	region.pitch_keycenter = pitch
	region.lokey = pitch
	region.hikey = tuples[index + 1][0] - 1  if index < 5 else pitch + 7
sfz.write(stdout)

The above script outputs:

// [Unnamed]

<global>
loop_mode=one_shot

<region>
lokey=40
hikey=44
sample=samples/40-E2.wav
pitch_keycenter=40

<region>
lokey=45
hikey=49
sample=samples/45-A2.wav
pitch_keycenter=45

<region>
lokey=50
hikey=54
sample=samples/50-D3.wav
pitch_keycenter=50

<region>
lokey=55
hikey=58
sample=samples/55-G3.wav
pitch_keycenter=55

<region>
lokey=59
hikey=63
sample=samples/59-B3.wav
pitch_keycenter=59

<region>
lokey=64
hikey=71
sample=samples/64-E4.wav
pitch_keycenter=64

SFZ structure and navigation.

The structure of an instance of SFZ follows the SFZ format. It may contain headers of various types, and headers may contain opcodes.

For programming covenience, the SFZ class extends the Header class, inheriting its attributes and methods.

The SFZ class exposes a "subheaders()" method which returns a list of headers which are immediate children of the SFZ. Subsequently, calling "subheaders()" on each subheader returns a list of headers contained in it. You could use this structure to iterate through the headers and opcodes in an SFZ.

Another way to iterate through is to call the "walk()" method. This is a generator function which recurses through the tree structure of the SFZ, yielding an element (header or opcode) on each iteration. The return value of "walk()" is a tuple of (element, depth).

For example:

from sfzen import SFZ

sfz = SFZ(filename)
for elem, depth in sfz.walk():
	if isinstance(elem, Header):
		print("  " * depth, elem)

The above example is redundant, however, as the "dump()" method does just this.

You can traverse up the hierarchy using the "parent" attribute of both Opcode and Header.

region = sample.parent

Opcodes and Opcode values

Opcodes may be instantiated and appended to a header's list of elements using this syntax:

opcode = region.append(Opcode("loop_mode", "one_shot"))

You could instantiate an Opcode and not append it, but unless the opcode is appended to the header, it isn't a part of the SFZ structure yet. This is particularly relevant when cloning opcodes (or headers) from different SFZs.

Individual opcodes can be retrieved using the "opcode(<opcode_name>)" method. This method returns an Opcode with the given name contained in the Header on which it is called:

opcode = region.opcode("loop_mode")

If the opcode is not defined in the header, the "opcode" method returns None.

To remove an opcode from a header, pass it to the header's "remove()" method:

region.remove(opcode)

You can also remove an opcode from a header by name, using the header's "remove_opcode(<opcode_name>)" method:

region.remove_opcode("loop_mode")

Get / set opcodes

An opcode can also be created and its value set by using the "set_opcode_value()" method:

region.set_opcode_value("loop_mode", "one_shot")

If the named opcode doesn't yet exist inside the header, it will be created.

The value of an opcode defined in any header can also be retrieved using the "get_opcode_value()" method:

loop_mode = region.get_opcode_value("loop_mode")

If the opcode is not defined in the header, the "get_opcode_value" method returns None.

Attribute access

The above "get_opcode_value" and "set_opcode_value" do almost the same thing as setting an opcode as an attribute of the header. In the example given above, the sample, pitch_keycenter, lokey, and hikey of each Region were all set using the syntax:

region.loop_mode = "one_shot"

When setting an attribute, and there is no opcode with the given opcode name contained in the header, it will be created in the header on which it is called.

The opcode value previously set can subsequently be retrieved by referencing the opcode name as an attribute of the Region:

loop_mode = region.loop_mode

Attribute access returns the "effective value" of the opcode; if there is no Opcode with the given "opcode_name" contained in the header, attribute lookup will move up the hierarchy to the parent header until it finds one, and return that value.

So if, for example, a <group> contains an opcode named "key", any <region> it contains will have the same value for key, unless otherwise defined in that region.

sfz = SFZ()
group = sfz.append(Group())
group.key = 48
region1 = group.append(Region())
region1.hivel = 63
region1.sample = "samples/48-pianissimo.wav"
region2 = group.append(Region())
region2.lovel = 64
region1.sample = "samples/48-forte.wav"

for region in group.regions():
	print(region.key, region.lovel, region.hivel)

Outputs:

key: 48, lovel: None, hivel: 63
key: 48, lovel: 64, hivel: None

Retrieving all opcodes

All of the Opcodes in a Header may be retrieved using the "opcodes()" method, which returns a dictionary. The keys are the opcode names, and the values are instances of the Opcode class.

all_opcodes = region.opcodes()

Inherited opcodes

To retrieve all the opcodes defined in a header, as well as all the opcodes defined all of it's ancestors in the SFZ tree, use the "inherited_opcodes()" method:

all_effective_opcodes = region.inherited_opcodes()

If "loop_mode" is not defined in the <region>, but is defined in a parent header, the "opcode()" method will not return it. If you want to return an opcode from a header and any of its parent headers, you can use the "iopcode()" method.

Like the "inherited_opcodes()" method described above, the "iopcode(<opcode_name>)" method will retrive an opcode defined in a parent header.

So, to retrieve the loop_mode which affects the sample, regardless of where it is defined, you will do:

effective_opcode = region.iopcode("loop_mode")

Aliases

Note that there are two opcodes in the SFZ format which define the way a sample is looped, "loopmode" and "loop_mode". You can use either one, as you prefer. Using attribute access, opcode aliases are also checked. So if you access the "loopmode" attribute of a <region> that has no Opcode named "loopmode", but does contain an Opcode named "loop_mode", the value of the Opcode named "loop_mode" will be returned for the "loopmode" attribute.

The following example should make this clear:

from sys import stdout
from sfzen import SFZ, Group, Region

sfz = SFZ()
group = sfz.append(Group())
region = group.append(Region())

print(region.loopmode)
# (region.loopmode returns "None")

group.loopmode = "loop_continuous"
print(region.loopmode)
# (region.loopmode returns "loop_continuous")

region.loopmode = "one_shot"
print(region.loopmode)
# (region.loopmode returns "one_shot")

print('-------')
sfz.write(stdout)

The output of the above script will be:

None
loop_continuous
one_shot
-------
// [Unnamed]

<group>
loopmode=loop_continuous

<region>
loopmode=one_shot

Opcode retrieval summary

Method	Returns	Inherited?	Follow aliases?
opcode	Opcode	No	No
iopcode	Opcode	Yes	No
get_opcode_value	value	No	No
.<attr>	value	Yes	Yes

Samples and regions

The "samples()" method of the Header class returns all the samples contained in that header, as well as all of it's descendant headers:

for sample in sfz.samples():
	region = sample.parent
	opcodes = region.opcodes()

Note that "samples()" is a generator function. To return a list, use "list(header.samples())".

The "<sample>" opcode points to a location in the filesystem, so filesystem -related methods are added to the Opcode class in the Sample class.

Two attributes of the Sample which might be of interest:

abspath (property): Returns (pathlib.Path) the absolute path, with symlinks resolved.

exists (method): Returns boolean True if file exists

Sample values are usually defined relative to the root path of the SFZ. To retrieve the absolute path to a sample file, use its "abspath" attribute:

for sample in sfz.samples():
	print(sample.abspath)

Normalization and validation

An opcode named "offset_cc22" will not be literally defined in the spec, but it is an instance of the "offset_ccN" opcode. To retrieve the opcode name defined in the spec, use the "normal_opcode()" function. Some functions, like "opcode_definition()" and "validator_for()" already normalize the opcode name when doing a lookup. Aliases are normalized as well.

Errors in parsing or opcode validation do not raise exceptions. SFZ errors which are detected during parsing are available in the SFZ "errors" attribute. This is a list of objects extending ValidationError. If no error occurred, this list will be empty.

An error detected during validation of an Opcode's value is available in the Opcode's "error" attribute. If no validation error occurred, this attribute will be None.

Validation of the SFZ structure occurs when an SFZ is instantiated and parsed. Opcode validation occurs not only during initial parsing, but whenever a value is set on an opcode.

Some opcodes accept any value above a minimum value, some accept any value between a minimum value and a maximum value, some accept only one value out of a selection of choices. The appropriate Opcode validator is chosen when the Opcode is instantiated, based on the opcode "value" definition found in the OPCODES dictionary (found in the sfzen.opcodes module). All of the validator classes extend the Validator class, which provides type checking as well.

The classes extending the ValidationError class include:

DataTypeError

The incorrect data type for the value was given. Since SFZ files are text files, all values are strings when first parsing. The validator checks that the string given can be coerced to the correct data type, and if not, generates this error.

MinError

The value given falls below the opcode's minimum value.

MaxError

The value given falls above the opcode's maximum value.

ChoiceError

The value given was not found in the list of valid choices.

NotFoundError

(Sample and Include) The referenced file was not found.

InvalidOpcodeError

An opcode was encountered with an unrecognized name.

InvalidHeaderError

A header was encountered with an unrecognized name.

Sample modes when using "save_as"

Normally, an SFZ file's sample paths are given as relative to the SFZ file. When saving an SFZ object, you have multiple choices as to how you want the samples saved and referenced. (This is particularly useful when copying an sfz.)

SAMPLES_ABSPATH

Sample values are changed to an absolute path to the original sample.

SAMPLES_RELPATH

Sample values are changed to a path relative to the original sample.

SAMPLES_COPY

The sample files are copied to a folder beneath the SFZ named "samples-<sfz_name>". Sample values are changed to a path relative to the new sfz path.

SAMPLES_SYMLINK

The sample files are symlinked in a folder beneath the SFZ named "samples-<sfz_name>". Sample values are changed to a path relative to the new sfz path.

SAMPLES_HARDLINK

The sample files are hardlinked in a folder beneath the SFZ named "samples-<sfz_name>". Sample values are changed to a path relative to the new sfz path.

SFZ simplification

After building up an SFZ, you can automatically create groups which define opcodes common to the regions they contain, using the "simplified()" method of the SFZ class. This creates a copy of the original SFZ, with some redundant opcodes removed and common opcodes grouped inside <group> and <global> headers.

To use the "simplified()" method in your code:

from sfzen import SFZ

source_sfz = SFZ('source.sfz')
simple = source_sfz.simplified()
simple.save_as('simple.sfz')

You can do the same as the above script using the command-line tool "sfz-copy", using the "--simplify" option.

Using simplification, I saw significant reductions on some .sfz files found in the public domain, as well as some generated from SoundFonts using Polyphone.

Space isn't really the issue, though. Readability is. After simplification, the unique charcteristics of each region stand out from the "noise" of all the other opcodes, which end up grouped in the global header.