soundbox-go, branch mainGo library to interface with soundbox deviceshttps://cgit.xengineering.eu/soundbox-go/atom2025-01-31T12:43:43ZUpdate CHANGELOG.md for version 0.2.22025-01-31T12:43:43Zxegineeringme@xegineering.eu2025-01-31T12:43:43Zurn:sha1:437c7ca98741bb178cedb9e940c6bcc14521d457Fix 100 % CPU usage2025-01-12T08:25:47Zxegineeringme@xegineering.eu2025-01-12T08:25:47Zurn:sha1:788b502b1ec4382558e9996aee5b6cb7a87f76e3
The `default` statement in the PipeWire reading caused 100 % CPU usage.
Introducing a one millisecond delay there fixes the issue.
Update CHANGELOG.md for version 0.2.12024-12-15T11:48:33Zxegineeringme@xegineering.eu2024-12-15T11:48:33Zurn:sha1:bde7e0d767e5e0b1072d05eff58bef16f5858e28pipewire: Assert OS is GNU / Linux2024-12-15T11:36:59Zxegineeringme@xegineering.eu2024-12-15T11:25:08Zurn:sha1:196ed013acad91cf89b7a1e086dbbe0d158166d3
PipeWire is a Linux-specific sound system. This commit adds a proper
error if the operating is not matching.
pipewire: Make sure only one instance is running2024-12-15T11:36:59Zxegineeringme@xegineering.eu2024-12-15T11:21:24Zurn:sha1:de2694f182790dd311265e378498d586f9b3ae66
The init / deinit of the PipeWire C library is not well handled for
multiple instances. Furthermore the PipeWire audio streaming relies on a
channel which is a global variable.
Both issues have to be fixed before multiple PipeWire capture instances
are allowed.
pipewire: Fix silence dropping2024-12-15T11:36:59Zxegineeringme@xegineering.eu2024-12-15T11:01:53Zurn:sha1:2184e703ba3c3fd79e792a78a70e76daa2aa1062
It used to drop silence per channel, not silence per sample. In the
unlikely case of perfect silence on one channel and sound on the other
this would switch effectively the channels.
pipewire: Implement capture tear-down2024-12-15T11:36:51Zxegineeringme@xegineering.eu2024-12-11T19:03:29Zurn:sha1:ca5ebc8a795114a72f4410f05b2270f24ead1d60
This closes the PipeWire process properly to not leak memory and remove
the PipeWire capture node of soundbox as soon as the context passed to
StreamPipewireContext() is closed.
pipewire: Disable auto-connect2024-12-14T15:52:55Zxegineeringme@xegineering.eu2024-12-14T15:52:55Zurn:sha1:62d3045df3283872628b0b8b8a8caeef1c226dfa
This used to connect the soundbox capture device directly to the default
input (usually system microphone). This is not wanted since the main
purpose of soundbox is not streaming the microphone but output of media
players.
Furthermore depending on the sound setup it could lead to a feedback
loop with a very loud noise.
pipewire: Avoid unnamed struct member init2024-12-10T20:28:00Zxegineeringme@xegineering.eu2024-12-10T20:28:00Zurn:sha1:5da49db791b6d70ca5ac4154dc49e55d6219ba33
This makes the code less readable.
Update CHANGELOG.md for release v0.2.02024-12-08T17:37:57Zxegineeringme@xegineering.eu2024-12-08T17:37:57Zurn:sha1:957d98544686d87d9df522b35981d1055b83c0b9Apply go fmt2024-12-08T17:36:10Zxegineeringme@xegineering.eu2024-12-08T17:36:10Zurn:sha1:477fb59944222966b8a9fc1d92893a336e827285Add examples to overview on generated API doc2024-12-08T17:34:11Zxegineeringme@xegineering.eu2024-12-08T17:31:56Zurn:sha1:f30e3668a6a65a8e5e43b3cc0b2030788b43857d
This makes it easier for readers to find them.
Add generate-compile-commands.sh2024-12-08T17:22:14Zxegineeringme@xegineering.eu2024-11-29T15:21:46Zurn:sha1:b76ac2daffb98fe7aea14385638150381cfb031b
This documents the `bear` call required to produce a proper
`compile_commands.json` here.
The JSON file is not committed since it contains absolute paths. It is
used for language servers like `clangd`.
pipewire: Fix huge latency by dropping silence2024-12-08T17:22:14Zxegineeringme@xegineering.eu2024-12-08T16:36:28Zurn:sha1:49f47e76a85b3477835a0adec05649507d94b2e8
The current architecture uses the following processing:
- capture raw audio from PipeWire as unsigned 16 bit integers
- convert with a `ffmpeg` process to OGG / FLAC
- stream the `ffmpeg` output to multiple soundboxes via TCP
Only the first part is different for URL sources. Since using PipeWire
significant latency (up to 15 seconds) were measured.
It turned out that this happens exactly when zero bytes (silence) are
fed into the `ffmpeg` process. This commit avoids this by dropping those
empty samples.
It has to be made sure that only samples are dropped where both channels
are zero. Otherwise audible noise is the result.
pipewire: Add experimental PipeWire support2024-12-08T17:22:10Zxegineeringme@xegineering.eu2024-11-17T11:51:15Zurn:sha1:06c4f8b0120f5598f9d179b8c0fea33df35659a8
This implements a PipeWire capture device which can be used as an input
source instead of the already available URL input.
Known issues with the current PipeWire support are:
- user has to connect the monitor of the default audio sink to the
capture device manually
- correct shutdown has to be tested
- multiple instances do not work
- medium code quality requires refactoring
Since this is nevertheless usable and possible unknown bugs should be
figured out in practise soon this implementation is already added.
Bugfixes and refactoring might follow.
Update CHANGELOG.md for version 0.1.52024-11-29T09:21:31Zxegineeringme@xegineering.eu2024-11-29T09:21:31Zurn:sha1:4513eb614707d129824e60270bd45cdee9f04d06Refactor streamContext()2024-11-29T09:16:57Zxegineeringme@xegineering.eu2024-11-29T09:16:57Zurn:sha1:f5f0d12b697e71e67a403791c1b93ffdf61e272a
This avoids an unnecessary additional for loop.
Handle internal error correctly2024-11-29T09:06:01Zxegineeringme@xegineering.eu2024-11-29T09:06:01Zurn:sha1:97af30e8fee2274a437a13d8c2c19e4bfe29ddeeFix variable buffer size2024-11-29T08:16:57Zxegineeringme@xegineering.eu2024-11-29T08:16:57Zurn:sha1:49dc5ae3c12f2c696302171b54b130d94dea05e0
The switch to `io.Copy()` to pump the data to the soundbox devices
removed the control over the buffer size of this copy action.
While for generic copy actions it is even an advantage when a big buffer
is used this is a problem for the soundbox use case. A big buffer used
during copy means that the first soundbox device gets audio data
significantly earlier than the later ones since `io.MultiWriter()` works
sequentially.
Thus this commit switches to `io.CopyBuffer()` where a buffer has to
provided. For that purpose the same buffer size is used as before the
refactoring to use `io.Copy()`.
Replace custom code by io package functions2024-11-28T21:44:28Zxegineeringme@xegineering.eu2024-11-28T21:44:28Zurn:sha1:99031596efbcf004188217d2a9e53ac5ce33b511
This makes use of two functions from this package:
- io.Copy()
- io.MultiWriter()
`io.Copy()` is used to move the data from whatever reader is provided.
`io.Multiwriter()` solves the issue that we need to stream to multiple
network connections at the same time (one for each soundbox).
Handle read error after processing buffer2024-11-27T20:55:40Zxegineeringme@xegineering.eu2024-11-27T20:55:40Zurn:sha1:bb84332ef7ca114f0c608bbea7f11451f010a831
This is recommended by the Go standard library. One reason is that a
Reader might deliver the last couple of bytes together with the EOF
error. This is only handled correctly if the returned bytes are
processed first and the error is handled later.
[1]: https://pkg.go.dev/io#Reader
Introduce io.Reader-based streamContext()2024-11-27T20:38:16Zxegineeringme@xegineering.eu2024-11-27T20:29:39Zurn:sha1:6b7da29eea100b92659a3f1f5df6958ebad544c8
This prepares the switch to adding more sources than web URLs.
Everything providing an io.Reader can then simply use this internal
function in the background to avoid code duplication.
Fix streaming only via first interface candidate2024-11-17T11:48:00Zxegineeringme@xegineering.eu2024-11-17T11:45:35Zurn:sha1:d4cf5ed8e38d1e5d0b8110e8959e002c216f073fUpdate CHANGELOG.md for version 0.1.42024-11-10T15:40:19Zxegineeringme@xegineering.eu2024-11-10T15:40:19Zurn:sha1:a95013aa0813d1644aa15780f360fa4dad2223a8Add CHANGELOG.md2024-11-10T15:36:26Zxegineeringme@xegineering.eu2024-11-10T15:36:26Zurn:sha1:6f51073ecb44f5476e37e3d5ab8d87e917116bae
This keeps track of the changes to this repository.
Avoid unnecessary variable for args2024-11-10T14:54:59Zxegineeringme@xegineering.eu2024-11-10T14:54:59Zurn:sha1:93d3ce8ca7d335fb15da64981321f82891c1493cUse Go code for output network stream2024-11-10T14:53:44Zxegineeringme@xegineering.eu2024-11-08T19:56:56Zurn:sha1:36a2fda37e837bc39c88a2fa38678b0f3bded041
Calling the external program `ffmpeg` should be avoided completely in
the future to make soundbox-go a pure Go code base. `ffmpeg` provides
the following functionality to soundbox-go:
- web radio input stream transport
- re-encoding of the audio stream
- output stream transport to soundbox devices
The last part should be replaced with this commit as a first step.
Rename cmd to args2024-11-08T19:35:24Zxegineeringme@xegineering.eu2024-11-08T19:35:24Zurn:sha1:e30a1b0768d35ec70e5d2c00e6ae87c0f267b881
This variable only contains the arguments for the called program. Thus
it should be named like this.
Fix outdated README.md2024-10-31T21:37:26Zxegineeringme@xegineering.eu2024-10-31T21:37:26Zurn:sha1:702c39c9bb3a6441de366aa30842960741017615Change module name and restructure content2024-10-31T21:10:36Zxegineeringme@xegineering.eu2024-10-31T21:10:36Zurn:sha1:79eeb90079e417f0a9d040c1de8f3278c628810f
The repository names for soundbox are named as below:
- app: soundbox-app
- Go library module: soundbox-go
- Device: soundbox
The Go module names were:
- app: xengineering.eu/soundbox/app
- Go library module: xengineering.eu/soundbox
This does not make clear which module is related to which repository
since the names are different. Thus it should be changed to:
- app: xengineering.eu/soundbox-app
- Go library module: xengineering.eu/soundbox-go
The import statement for the library is then:
import "xengineering.eu/soundbox-go/soundbox"
This is a bit longer but it keeps the property that the library is
referenced inside the code by the simple name `soundbox`.