Native high-fidelity stereo voice, inside Discord.
A standalone Audiocord build with the HD voice engine baked in - no file swapping.
End-to-end high-fidelity audio processing, tuned within Discord's official boundaries.
End-to-end 24-bit / 96 kHz float32 audio processing with near-lossless Opus HD encoding at 530 kbps.
DRED v2 greatly improves robustness to packet loss by embedding very low-bitrate redundancy into the Opus stream.
225-channel ambisonics (14th-order) with ACN/SN3D encoding.
Continuously improved with new features, optimizations, and compatibility refinements.
Neural reconstruction of high-frequency speech (8-20 kHz) from wideband input.
Legacy clients receive high-quality 48 kHz / 510 kbps streams.
Discord enforces a hard 48 kHz sample-rate cap and mono/stereo channel limits. Opus 1.6+ HD encoding (up to 530 kbps) is supported, but end-to-end fidelity is limited by Discord's infrastructure. This is an optimisation within official boundaries.
Audiocord exposes the full voice pipeline in-app. Every setting below is implemented and adjustable from the client.
Drive the Opus encoder well past Discord's defaults, with a slider capped at 530 kbps and a min-bitrate floor to hold quality under packet loss.
True two-channel stereo encoding baked into the client - no userscript, no SDP munging. Flip it on and the encoder sends real stereo.
Full-band 48 kHz audio path, the maximum Discord's infrastructure accepts, with the libopus 1.6.1 encoder running fullband mode.
Toggle the QEXT residual extension, lock constant bitrate or run VBR, and set CELT-only mode for the cleanest high-bitrate music path.
Force in-band FEC for packet-loss resilience, disable DTX to kill silence cutoffs, and push encoder complexity to 10 for maximum fidelity.
A built-in input meter so you can see your levels in real time and confirm the encoder is receiving signal before you transmit.
Encrypted message backups are sealed end-to-end with a hybrid X25519 + ML-KEM-768 scheme, so they stay private even against future quantum attacks - only you hold the keys.
Adjust packet time (minptime/maxptime/ptime) to trade latency against efficiency, tuned for the high-bitrate path.
Audiocord changes how your own client encodes and sends audio. It does not change what Discord's servers deliver to other people - here is the precise picture.
Audiocord changes the local WebRTC / libopus encoder path before audio leaves your client.
The userscript writes custom Opus fmtp values (qext, cbr, usedtx, useinbandfec, maxaveragebitrate, minptime, maxptime, ptime, complexity) and the patched native WebRTC Opus encoder reads them.
The local libopus build has ENABLE_QEXT, WebRTC has enable_qext, and qext=1 calls OPUS_SET_QEXT(1) - real local QEXT support.
The WebRTC Opus cap is raised locally and QEXT raises the internal packet-size budget.Caveat: Discord's SFU can still cap, drop or mangle anything beyond its accepted envelope.
CBR, DTX, ptime, CELT-only, complexity, QEXT and bitrate changes alter how the local encoder produces packets.Caveat: on a normal Discord voice channel it is still RTP/SRTP through Discord's SFU.
Two ways to get HD voice: the standalone Audiocord client, or the manual Node file-swap method. Here is how they compare.
The Node method means manually swapping discord_voice.node into Discord. Audiocord is the standalone client with the same HD voice engine baked in, so there is nothing to swap and nothing to reapply after Discord updates.
| Feature | Node (file-swap) | Audiocord |
|---|---|---|
| Installation | Manual file replacement | Standalone installer |
| Survives Discord updates | No - reapply each update | Yes - self-contained |
| Stereo support | Full | Full |
| Bitrate control | 512 kbps lock + floor + GoogCC NOP | In-app, up to 530 kbps |
| Voice engine | libopus 1.3 (patched discord_voice.node) | libopus 1.6.1 |
| Native stereo | Forced on (no toggle) | Toggle in-app |
| Platform support | Win/Linux | Windows |
| Maintenance status | Slower - Audiocord is taking over | Active - primary focus |
All critical voice, bitrate, network, FEC and video offsets discovered and applied.
BWE audio/video params, CwndPushback, forced FEC and a min-bitrate floor prevent degradation.
Regularly updated for new Discord versions (app-1.0.9226+) with active development.
MIT-licensed, transparent codebase on Codeberg with community contributions.
Enable near-lossless HD audio in Discord with the custom voice module.
Grab the files from the official repository:
index.js (optional)discord_voice.node (required)ffmpeg.dll (legacy clients)Press Win + R, paste the path below and press Enter:
This opens the main Discord folder where app versions are stored.
Open the latest app folder, then navigate to:
Fully quit Discord first: right-click the tray icon → Exit Discord.
Drop the downloaded files into that folder, overwriting:
index.jsdiscord_voice.nodeReopen Discord and join a voice channel as usual.
⚠️ Future Discord updates may overwrite these files - reapply after each major update.
Open, independent and ad-free. Live project news and our Discord, side by side.
Anonymous, aggregate usage only. No personal data is sold or shared, and you can opt out at any time in the panel below.
Free, open-source Android apps built alongside Audiocord - the same focus on real audio quality and doing things natively, no root or hacks.
A community fork of Kizzy (by dead8309) - a Discord Rich Presence manager for Android. Set a custom status, or show your now-playing track or the app you are using, straight on your Discord profile. Built in Kotlin / Jetpack Compose and installs alongside the original.
codeberg.org/UnpackedX/Kizzey ↗A real-time stereo microphone monitor with a full on-device effects rack - low-cut filter, noise gate, EQ, compression and more. Hear yourself live, or send a clean processed mic feed to a PC or speakers over Bluetooth A2DP. No root and no virtual-mic hacks.
codeberg.org/UnpackedX/AudioMon ↗Two ways to get HD voice: the standalone Audiocord client, or the manual Node file-swap method. Browser and Ripcord options are also available.
Standalone client with a patched libopus 1.6.1 voice engine. Native stereo toggle, configurable bitrate and 48 kHz.
🔒The 4th May node. Full stereo, 512 kbps, 28 patches. Needs the corruption fix below.
→Resolves the audio-corruption issue. Replaces the previous index.js. Apply alongside the node above.
Corruption issue discovered by ch4n3lx_. →Userscript for Discord Web with a GUI for bitrate, sample rate, network options and a debug menu.
→libopus 1.6.1 with QEXT residual layer. CELT-only locked, lsb_depth pinned to 24-bit. ~819 kbps wire.
🔒The 24-bit opus.dll is distributed exclusively through our Discord during early access. Join the server and open #audio-patches for the latest build.
Ripcord 24-bit builds are in early access. Enter the invite code from Discord to unlock the download.
Audiocord is in early access. Enter the invite code from #audiocord, or join the Discord to get one.
Click the strip to set a custom accent colour.
A live WebGL gradient replaces the static aurora. Turn off to save battery/GPU.
Aggregate page views only, no personal data. Stored as a local preference; turn off to opt out.
Uses spoo.me for anonymous download counts. Turn off to use direct raw links from the link map.