Writing audio to disk

[dayunbao]

I'm trying to save the audio from the playback of a sequenced track. I'm keeping everything in groups to make sure the order of execution is correct:

self.instruments_group = self.server.add_group()
self.effects_group = self.instruments_group.add_group(add_action=AddAction.ADD_AFTER)
self.tracks_group = self.effects_group.add_group(add_action=AddAction.ADD_AFTER)

This is what I see when dumping the tree immediately after triggering sample playback:

NODE TREE 0 group
    1 group
        1000 group
            1020 sample_player
                drum_buff: 0.0, out_bus: a16
        1001 group
            1003 reverb
                damping: 0.5, in_bus: a16, mix: 0.33, out_bus: 0.0, room_size: 0.5
        1002 group
            1012 audio_to_disk
                buffer_number: 12.0, in_bus: 0.0

I have the sample_player's output routed through reverb, and reverb's output going to the standard output (0).

This is what the DiskOut SynthDef looks like:

@synthdef()
def audio_to_disk(in_bus, buffer_number):
    input = In.ar(bus=in_bus, channel_count=2)
    DiskOut.ar(buffer_id=buffer_number, source=input)

When I start playing back a sequenced track, I'm calling this method:

def start_recording_to_disk(self) -> None:
        self.recording_buffer = self._create_buffer()
        self.recording_buffer.write(
            file_path=self.buffer_file_path,
            header_format='WAV',
            leave_open=True,
        )

        self.audio_to_disk_synth = self.group.add_synth(
            synthdef=audio_to_disk,
            buffer_number=self.recording_buffer.id_,
            in_bus=self.recording_bus,
        )

The _create_buffer method:

def _create_buffer(self) -> Buffer:
    return self.server.add_buffer(
        channel_count=self.BUFFER_CHANNELS, # 2
        frame_count=self.BUFFER_FRAME_COUNT, # 262144
    )

When playback stops, I'm calling this method:

def stop_recording_to_disk(self) -> None:
    self.audio_to_disk_synth.free()
    self.recording_buffer.close()
    self.recording_buffer.free()

However, I'm not getting any audio recorded. I'm stumped. All the buses look correct to me, as does the order of groups. Any suggestions?

[josephine-wolf-oberholtzer]

Ingredients generally look good. Might need to put the buffer freeing into the completion of closing the buffer.

[josephine-wolf-oberholtzer]

To clarify, /b_close is an async command, so might be that you're starting the close process, then freeing the buffer before closing finishes.

Something like this should wait to free the buffer until closing finishes:

self.recording_buffer.close(on_completion: lambda ctx: self.recording_buffer.free())

And if it's not that, I'm not sure. Would need an MRE to play with.

[dayunbao]

Here's an MRE. It isn't working for me.

import time
from pathlib import Path

from supriya import synthdef
from supriya import AddAction, Buffer, Server, Synth
from supriya.ugens import Envelope, EnvGen, In, LFSaw, Limiter, Out, Pan2
from supriya.ugens.diskio import DiskOut

@synthdef()
def audio_to_disk(in_bus, buffer_number):
    input = In.ar(bus=in_bus, channel_count=2)
    DiskOut.ar(buffer_id=buffer_number, source=input)

@synthdef()
def saw(frequency=440.0, amplitude=0.5, gate=1, out_bus=0):
    signal = LFSaw.ar(frequency=[frequency, frequency - 2])
    signal *= amplitude
    signal = Limiter.ar(duration=0.01, level=0.1, source=signal)
    adsr = Envelope.adsr()
    env = EnvGen.kr(envelope=adsr, gate=gate, done_action=2)
    signal *= env
    signal = Pan2.ar(source=signal, position=0.0, level=1.0)
    Out.ar(bus=out_bus, source=signal)

def create_buffer(server: Server) -> Buffer:
    buffer = server.add_buffer(
        channel_count=2,
        frame_count=262144,
    )

    server.sync()

    return buffer

def create_buffer_file_path() -> Path:
    dir_path = Path(__file__).parent / 'recordings'
    if not dir_path.exists():
        dir_path.mkdir()

    file_path =  dir_path / 'recording.wav'
    
    if not file_path.exists():
        file_path.touch()
    
    return file_path

def main() -> None:
    server = Server().boot()
    server.add_synthdefs(audio_to_disk, saw)
    server.sync()
    saw_synth = start_playing_synth(server=server)
    file_path = create_buffer_file_path()
    recording_buffer = create_buffer(server=server)
    start_writing_buffer(recording_buffer=recording_buffer, file_path=file_path)
    audio_to_disk_synth = start_writing_audio_to_disk(server=server, recording_buffer=recording_buffer)
    print(server.dump_tree())
    time.sleep(5)
    audio_to_disk_synth.free()
    saw_synth.set(gate=0)
    saw_synth.free()
    recording_buffer.close(on_completion=lambda _: recording_buffer.free())

def start_playing_synth(server: Server) -> Synth:
    saw_synth: Synth = server.add_synth(
        synthdef=saw,
        in_bus=0,
    )

    return saw_synth

def start_writing_audio_to_disk(server: Server, recording_buffer: Buffer) -> Synth:
    audio_to_disk_synth: Synth = server.add_synth(
        synthdef=audio_to_disk,
        buffer_number=recording_buffer.id_,
        in_bus=0,
        add_action=AddAction.ADD_TO_TAIL,
    )

    return audio_to_disk_synth

def start_writing_buffer(recording_buffer: Buffer, file_path: Path) -> None:
    recording_buffer.write(
        file_path=file_path,
        header_format='WAV',
        leave_open=True,
    )


if __name__ == '__main__':
    main()

[josephine-wolf-oberholtzer]

Biiig thank you for the MRE. Super helpful. The problem is a little bit of both of us.

1. I need to apply the following changes to make sure the /b_write command says the frame_count is 0, not -1:

diff --git a/supriya/contexts/core.py b/supriya/contexts/core.py
index cfd3e0d3..45c4ac58 100644
--- a/supriya/contexts/core.py
+++ b/supriya/contexts/core.py
@@ -1352,11 +1352,11 @@ class Context(metaclass=abc.ABCMeta):
         request = WriteBuffer(
             buffer_id=buffer.id_,
             path=file_path,
-            frame_count=frame_count or -1,
+            frame_count=frame_count if frame_count is not None else -1,
             header_format=header_format,
             leave_open=leave_open,
             sample_format=sample_format,
-            starting_frame=starting_frame or 0,
+            starting_frame=starting_frame if starting_frame is not None else 0,
         )
         return self._add_request_with_completion(request, on_completion)

2. You need to wait for the /b_close to complete before your script exits. Add a final server.sync() after you close the buffer.

I'll have a PR up for the Context.write_buffer() change momentarily.

[josephine-wolf-oberholtzer]

#457

[dayunbao]

👏

…

On Fri, Feb 28, 2025, 7:19 PM Joséphine Wolf Oberholtzer < ***@***.***> wrote: #457 <#457> — Reply to this email directly, view it on GitHub <#456 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVNNEUE2NHK4XUSKC5WLJT2SCSB5AVCNFSM6AAAAABYCGBOJGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMZVGQ4TSMY> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[josephine-wolf-oberholtzer]

@dayunbao Just released the fix on PyPI: https://pypi.org/project/supriya/25.2b2/

[dayunbao]

Awesome! Thanks!

[josephine-wolf-oberholtzer]

You're welcome. And thanks again for the MRE. I can move very fast when I have good examples to work against.

[dayunbao]

I'll try to start with those in the future.

…

On Fri, Feb 28, 2025, 8:13 PM Joséphine Wolf Oberholtzer < ***@***.***> wrote: You're welcome. And thanks again for the MRE. I can move very fast when I have good examples to work against. — Reply to this email directly, view it on GitHub <#456 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVNNEQKEDEDLPSRKNRRCQ32SCYNZAVCNFSM6AAAAABYCGBOJGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMZVGU3TAOA> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[dayunbao]

Seems that I've found another issue. When I create the buffer, the value for frame_count is introducing leading silence. I mean this code:

buffer = server.add_buffer(
        channel_count=2,
        frame_count=65536,
    )

When frame_count equals 65536, it adds roughly one second of silence at the beginning of the recording. When it's equal to 262144, it adds about 5 seconds. Did you experience this? Setting the start_frame to zero in buffer.write doesn't change anything. It seems like the buffer is starting to write after the specified frame_count. I'm not sure if that's expected behavior.

import sys
import time
from pathlib import Path

from supriya import synthdef
from supriya import AddAction, Buffer, Server, Synth
from supriya.ugens import Envelope, EnvGen, In, LFSaw, Limiter, Out, Pan2
from supriya.ugens.diskio import DiskOut

@synthdef()
def audio_to_disk(in_bus, buffer_number):
    input = In.ar(bus=in_bus, channel_count=2)
    DiskOut.ar(buffer_id=buffer_number, source=input)

@synthdef()
def saw(frequency=440.0, amplitude=0.5, gate=1, out_bus=0):
    signal = LFSaw.ar(frequency=[frequency, frequency - 2])
    signal *= amplitude
    signal = Limiter.ar(duration=0.01, level=0.1, source=signal)
    adsr = Envelope.adsr()
    env = EnvGen.kr(envelope=adsr, gate=gate, done_action=2)
    signal *= env
    signal = Pan2.ar(source=signal, position=0.0, level=1.0)
    Out.ar(bus=out_bus, source=signal)

def create_buffer(server: Server) -> Buffer:
    buffer = server.add_buffer(
        channel_count=2,
        frame_count=65536, # <- Change me to 262144 to see the difference
    )

    server.sync()

    return buffer

def create_buffer_file_path() -> Path:
    dir_path = Path(__file__).parent / 'recordings'
    if not dir_path.exists():
        dir_path.mkdir()

    file_path =  dir_path / 'recording.wav'
    
    if not file_path.exists():
        file_path.touch()
    
    return file_path
    

def main() -> None:
    server = Server().boot()
    server.add_synthdefs(audio_to_disk, saw)
    server.sync()
    file_path = create_buffer_file_path()
    recording_buffer = create_buffer(server=server)
    start_writing_buffer(recording_buffer=recording_buffer, file_path=file_path, server=server)
    audio_to_disk_synth = start_writing_audio_to_disk(server=server, recording_buffer=recording_buffer)
    saw_synth = start_playing_synth(server=server)
    time.sleep(5)
    audio_to_disk_synth.free()
    saw_synth.set(gate=0)
    saw_synth.free()
    recording_buffer.close(on_completion=lambda _: recording_buffer.free())
    server.sync()
    sys.exit(0)


def start_playing_synth(server: Server) -> Synth:
    saw_synth: Synth = server.add_synth(
        synthdef=saw,
        in_bus=0,
    )

    return saw_synth

def start_writing_audio_to_disk(server: Server, recording_buffer: Buffer) -> Synth:
    audio_to_disk_synth: Synth = server.add_synth(
        synthdef=audio_to_disk,
        buffer_number=recording_buffer.id_,
        in_bus=0,
        add_action=AddAction.ADD_TO_TAIL,
    )

    return audio_to_disk_synth

def start_writing_buffer(recording_buffer: Buffer, file_path: Path, server: Server) -> None:
    recording_buffer.write(
        file_path=file_path,
        header_format='WAV',
        leave_open=True,
        starting_frame=0,
    )

    server.sync()


if __name__ == '__main__':
    main()

[josephine-wolf-oberholtzer]

Make sure you pass frame_count=0 in recording_buffer.write(...). Otherwise it defaults to frame_count=-1 which will write all of the (zeroed) samples in the buffer to disk before it gets to writing what your synths have written into it.

It can be really helpful to validate what commands your system is sending by wrapping blocks of code in with server.osc_protocol.capture() as transcript: and then dumping out what OSC commands you actually sent, and compare them against what's in the official Server Command Reference.

def main() -> None:
    server = Server().boot()
    server.add_synthdefs(audio_to_disk, saw)
    server.sync()
    file_path = create_buffer_file_path()
    with server.osc_protocol.capture() as transcript:
        recording_buffer = create_buffer(server=server)
        start_writing_buffer(
            recording_buffer=recording_buffer, file_path=file_path, server=server
        )
        audio_to_disk_synth = start_writing_audio_to_disk(
            server=server, recording_buffer=recording_buffer
        )
        saw_synth = start_playing_synth(server=server)
        time.sleep(5)
        audio_to_disk_synth.free()
        saw_synth.set(gate=0)
        saw_synth.free()
        recording_buffer.close(on_completion=lambda _: recording_buffer.free())
        server.sync()
    for x in transcript.filtered(sent=True, received=False, status=False):
        print(repr(x))
    sys.exit(0)

Before I explicitly added the frame_count=0 I saw:

➜ python discussion456.py
OscMessage('/b_alloc', 0, 65536, 2)
OscMessage('/sync', 2)
OscMessage('/b_write', 0, '/Users/josephine/Source/github.com/supriya-project/supriya/recordings/recording.wav', 'wav', 'int24', -1, 0, 1)
OscMessage('/s_new', 'audio_to_disk', 1000, 1, 1)
OscMessage('/s_new', 'saw', 1001, 0, 1)
OscMessage('/n_free', 1000)
OscMessage('/n_set', 1001, 'gate', 0.0)
OscMessage('/n_set', 1001, 'gate', 0.0)
OscMessage('/b_close', 0, OscMessage('/b_free', 0))
OscMessage('/sync', 3)

And after I added that keyword argument I got:

➜ python discussion456.py
OscMessage('/b_alloc', 0, 65536, 2)
OscMessage('/sync', 2)
OscMessage('/b_write', 0, '/Users/josephine/Source/github.com/supriya-project/supriya/recordings/recording.wav', 'wav', 'int24', 0, 0, 1)
OscMessage('/s_new', 'audio_to_disk', 1000, 1, 1)
OscMessage('/s_new', 'saw', 1001, 0, 1)
OscMessage('/n_free', 1000)
OscMessage('/n_set', 1001, 'gate', 0.0)
OscMessage('/n_set', 1001, 'gate', 0.0)
OscMessage('/b_close', 0, OscMessage('/b_free', 0))
OscMessage('/sync', 3)

Note the /b_write command's last few arguments go from -1, 0, 1 to 0, 0, 1. We want that -1 which represents "number of frames to write" to be 0 instead so we don't write out the full empty buffer before continuing to write the rest of the incoming audio. I'd link the SCR section, but they don't have anchors on the page so just search for /b_write there.

[dayunbao]

Ah, yup. That was the problem. Thanks so much! Good advice. I'll start doing that from now on.

…

On Sat, Mar 1, 2025, 4:52 PM Joséphine Wolf Oberholtzer < ***@***.***> wrote: Make sure you pass frame_count=0 in recording_buffer.write(...). Otherwise it defaults to frame_count=-1 which will write all of the (zeroed) samples in the buffer to disk before it gets to writing what your synths have written into it. It can be really helpful to validate what commands your system is sending by wrapping blocks of code in with server.osc_protocol.capture() as transcript: and then dumping out what OSC commands you actually sent, and compare them against what's in the official Server Command Reference <https://doc.sccode.org/Reference/Server-Command-Reference.html>. def main() -> None: server = Server().boot() server.add_synthdefs(audio_to_disk, saw) server.sync() file_path = create_buffer_file_path() with server.osc_protocol.capture() as transcript: recording_buffer = create_buffer(server=server) start_writing_buffer( recording_buffer=recording_buffer, file_path=file_path, server=server ) audio_to_disk_synth = start_writing_audio_to_disk( server=server, recording_buffer=recording_buffer ) saw_synth = start_playing_synth(server=server) time.sleep(5) audio_to_disk_synth.free() saw_synth.set(gate=0) saw_synth.free() recording_buffer.close(on_completion=lambda _: recording_buffer.free()) server.sync() for x in transcript.filtered(sent=True, received=False, status=False): print(repr(x)) sys.exit(0) Before I explicitly added the frame_count=0 I saw: ➜ python discussion456.py OscMessage('/b_alloc', 0, 65536, 2) OscMessage('/sync', 2) OscMessage('/b_write', 0, '/Users/josephine/Source/github.com/supriya-project/supriya/recordings/recording.wav', 'wav', 'int24', -1, 0, 1) OscMessage('/s_new', 'audio_to_disk', 1000, 1, 1) OscMessage('/s_new', 'saw', 1001, 0, 1) OscMessage('/n_free', 1000) OscMessage('/n_set', 1001, 'gate', 0.0) OscMessage('/n_set', 1001, 'gate', 0.0) OscMessage('/b_close', 0, OscMessage('/b_free', 0)) OscMessage('/sync', 3) And after I added that keyword argument I got: ➜ python discussion456.py OscMessage('/b_alloc', 0, 65536, 2) OscMessage('/sync', 2) OscMessage('/b_write', 0, '/Users/josephine/Source/github.com/supriya-project/supriya/recordings/recording.wav', 'wav', 'int24', 0, 0, 1) OscMessage('/s_new', 'audio_to_disk', 1000, 1, 1) OscMessage('/s_new', 'saw', 1001, 0, 1) OscMessage('/n_free', 1000) OscMessage('/n_set', 1001, 'gate', 0.0) OscMessage('/n_set', 1001, 'gate', 0.0) OscMessage('/b_close', 0, OscMessage('/b_free', 0)) OscMessage('/sync', 3) Note the /b_write commands last few arguments go from -1, 0, 1 to 0, 0, 1. We want that -1 which represents "number of frames to write" to be 0 instead so we don't write out the full empty buffer before continuing to write the rest of the incoming audio. I'd link the SCR section, but they don't have anchors on the page so just search for /b_write there. — Reply to this email directly, view it on GitHub <#456 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVNNEQ7L7DJFV3PE7C5AW32SHJVBAVCNFSM6AAAAABYCGBOJGVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMZWGA4TSNA> . You are receiving this because you were mentioned.Message ID: ***@***.*** com>

[josephine-wolf-oberholtzer]

And, like, philosophically, Supriya has no power over SuperCollider beyond what arguments are used to boot the scsynth process, and what OSC commands are sent (and when). So the real source of truth for "what happened and why" is always "what commands were sent, what responses were received, and in what order".

[josephine-wolf-oberholtzer]

Everything that you interact with in Supriya is simply an abstraction / orchestration / convenience layer over OSC.