Audio

Inventory of the audio layer for both games. All figures originate from tools/probe_audio.py, which runs against the original data in base/_wolfsdk_backup (the installation is patched). Nothing was written, no game was launched.

Interpretation. A Magic-Byte that was read appears here as a Finding. Anything inferred solely from names, sizes, or analogy appears as a Hypothesis and is marked accordingly.

0. Executive Summary

The short path to "louder and brasher" does not go through new audio files, but rather through three text files: sound:inheritences/master (overwrites 9364 of 9444 Sound Decls), soundmixing:default (13 Mixbuses with gain), and family nodes such as inheritences/weapons/weapons_pl.

Update 2026-10-04. The TNC column is outdated: a TNC sound replacement is built (sections 9–11); held back in WolfSDK 0.2.0, enabled since 0.3.0 (not yet tried in game).

1. Which Asset Types Exist

Of 105 Asset Types and 203,843 Assets, seven are audio-related:

The task-named type names soundshader, snd, music, voice, reverb, and soundmaterial do not exist in TNO. What they mean content-wise is located elsewhere:

• soundshader → this is exactly the type sound; the fields are internally named soundShaderFlags.
• music → not a type, but a namespace: 225 sound Decls under music/…, e.g., music/menu/main_menu, music/c01p1/mus_c01p1_crashlanding, music/journal/house_of_the_rising_sun.
• voice → voiceover + voicetrack.
• reverb → not a type. Reverb is a) a namespace in Sound Decls (weapons/handgun/player/reverb/dw_large_room, …/dw_outdoor, …_rear) and b) the fields wetFalloff, wetVolume, wetMaxDistanceOffset. The Asset Type env (627 items) is Image Environment (Cubemaps, Color Grading), not Reverb — verified against env:default.
• soundmaterial → does not exist; material references run via impactSounds Decls (metalSound, stoneSound, fleshSound, woodSound, cardboardSound, …).

2. The Actual Data Format

Magic Bytes, all read (Finding):

No RIFF/WAVE, no XMA, no MSADPCM, no Bink Audio (BCF/KB2) in the audio paths. The .bik files under base/bink/ are videos with their own audio track and do not belong to the Sound Engine.

The filenames in the Decls end with .wav — this is the source name from production, not the delivered format. The game uses Ogg Vorbis.

bsnf — the Sample Descriptor

There are two variants, both exactly 74 or 272 bytes in size (no other sizes occur: 33.610 × 74 B, 27.708 × 272 B).

Monolingual, 74 Bytes:

u8[4]    "bsnf"
BE u32   Number of language entries (= 1)
u8[16]   Null bytes
BE u32   @24  Length of the Ogg stream in bytes
BE u32   @28  32              (meaning unconfirmed)
BE u32   @32  Hash/ID
BE u32   @36  0
BE u32   @40  Granule = Number of PCM samples
BE u32   @44  0
BE u32   @48  Length, repetition of @24
--- starting at @52 a WAVEFORMATEX in Little Endian ---
LE u16   wFormatTag        0x674F = Ogg Vorbis Mode 1
LE u16   nChannels         1 or 2
LE u32   nSamplesPerSec    48000
LE u32   nAvgBytesPerSec   192000 (stereo) / 96000 (mono)
LE u16   nBlockAlign       4 / 2
LE u16   wBitsPerSample    16
LE u32   Granule, repetition of @40

Multilingual, 272 Bytes: "bsnf", BE u32 = 4, followed by four 24-byte headers (char[16] language name, BE u32 length, BE u32 offset of the associated 42-byte sub-block within this descriptor), then the four 42-byte blocks starting at offset 104. Observed languages: english, french, italian, spanish.

Example sample:sound/projectile/bullet_impacts/concrete/imp_concrete_01.wav (path generated/sound/…/imp_concrete_01_vorbis.bsnd): Length 17.619 B, Granule 38.822 Samples → 0.81 s, Stereo, 48 kHz.

`"VTR" — the Voicetrack Descriptor

voicetrack:voicetrack/english/ai/efforts/eff_soldier_meleehit_01.vtr is 1.344 bytes in size and contains: Magic 22 56 54 52, BE u32 = 4, four 24-byte language headers (english/french/italian/spanish, each 0x136 = 310 bytes, offsets 0, 414, 724, 1034), then four 310-byte blocks with Magic 22 52 54 56 (`"RTV"). 104 + 4 × 310 = 1.344 — the math adds up exactly. Also no audio data here, only description. Hypothesis: the 310-byte block contains lip-sync/viseme data, because it is too large for simple pointers and the blocks are equal in length per language.

3. Where the actual audio bytes lie — the critical finding

They do not reside in the chunkN.resources. The archives contain only the 74/272-byte descriptors. The Ogg streams are located in two containers that are not listed in master.index:

base/streamed.resources   1.157.177.472 B   begins with "OggS"
base/english.streamed       415.137.488 B   begins with "OggS"

They are addressed via the variable remainder of the index entry — that section which docs/formats.md describes as "longer for audio types, not fully understood". 20 bytes behind the offset/usize/csize triplet stands a BE u32: the byte offset into the container.

Verification across all 61,318 sample entries: for each one, the Ogg page chain was traversed from the indicated offset to the EOS page, and the measured length was compared against the length from the bsnf.

streamed.resources    33,455 entries, length matches exactly
english.streamed      27,863 entries, length matches exactly
                           0 deviations

The container assignment is by name: everything under sound/vo/english/… lies in english.streamed, everything else in streamed.resources.

Slot Grid

streamed.resources   14,454 occupied slots, 14,454 of them on the 16384-byte grid (100 %)
                     free space per slot: min 0, median 8,607, max 3,144,537 B
english.streamed      5,091 occupied slots,  5,091 of them on the 16384-byte grid (100 %)
                     free space per slot: min 0, median 8,041, max 16,380 B

No single overlap. Each sound receives ceil(Länge / 16384) * 16384 bytes. There are fewer slots than entries because the same samples are indexed in multiple chunks.

Answer to the Core Question

The question was: normal archive entries or named resource blocks with pre-calculated size (such as the particles, which failed with idFile_RingBuffer file '<name>.bin' skipping <n> bytes)?

Neither nor — a third case, and it is cheaper than both.

The structural difference from particles is measurable: particle has two entries per asset, 2.672 .decl and 2.672 .bin under the same name. The .bin (Magic 06 54 52 50) is the compiled particle — that is the named resource block at which all particle mods failed. sound has no .bin counterpart. Its .decl entries carry the same short entry suffix (9 zeros + sequential BE-u32-ID) as entityDef, weapon, and material — exactly the types this SDK already successfully patches.

For audio bytes: they are not an archive entry at all, but a piece of file data at a fixed address with a length field located in the archive. The length is therefore changeable, as long as the slot accommodates it.

4. The sound Decls: the simple Modding Path

9,445 unique sound-Decl names, plaintext, fully readable by the parser in wolfsdk/decl.py. Namespaces: vo 5157, anim_events 1228, ambience 676, weapons 501, enemies 468, movements 272, music 225, footsteps 223, explosions 44, impacts 24.

The Fields Asked For

Random selection specifics: 1,717 Decls draw from more than one file (Distribution: 1 file 7614x, 3 files 316x, 5 x252, 6 x175, 4 x166, 8 x158). Examples:

sound:projectile/bullet_impacts/imp_flesh      num = 15   (Body hits)
sound:weapons/handgun/player/handgun_46_fire   num =  9
sound:projectile/bullet_impacts/imp_metal      num = 14

A complete example, verbatim from the archive:

{
    inherit = "inheritences/weapons/weapons_pl";
    edit = {
        parms = {
            minDistance = 50;
            falloff = "sound/falloffs/slow";
            wetFalloff = "sound/falloffs/slow";
            frontDirectionalAngle = 25;
        }
        soundFiles = {
            num = 9;
            item[0] = "sound/Weapons/handgun/handgun_fire_01.wav";
            …
            item[8] = "sound/Weapons/handgun/handgun_fire_09.wav";
        }
        rmsTable = "sound/hdr_curves/weapons/handgun/handgun_fire_05";
    }
}

This means: custom weapon sounds without new audio data are possible by redirecting item[i] to an existing sample — such as placing the pistol on weapons/m60/… or increasing/decreasing num to alter the spread. The sample name must point to an existing sample asset.

Inheritance — here lies the lever

99.1% of all Sound Decls (9,364 out of 9,445) inherit transitively from one file:

sound:inheritences/master
{
    edit = {
        parms = {
            volume = {
                offset = 10;
            }
            lfeVolume = -60;
        }
    }
}

A single number, 97 bytes, and everything gets louder. Below that are 105 family nodes targeting groups specifically:

For "brutal impacts," these are concrete: inheritences/impacts (raise volume.min, increase maxShaders from 2, so multiple impacts remain audible simultaneously) and sound:projectile/bullet_impacts/imp_flesh.

soundmixing — the mixing console level

26 Decls, soundmixing:default is the base (4,163 bytes), others inherit and override per chapter (c01p1, c01p2, …). Master bus with Limiter (threshold = -0.2) and 13 sub-buses with gain in dB, each with Compressor and Sidechain on SSB_BUS8:

subBus[ 0] World Sfx Master              gain = -9
subBus[ 1] Player Weapon                 gain = -11
subBus[ 2] Music: Scripted               gain = -6
subBus[ 3] Music: System                 gain = -6
subBus[ 4] Ambience                      gain = -4
subBus[ 5] Explosions                    gain = -5
subBus[ 6] Special Enemy Voice           gain = -5
subBus[ 7] Special Enemy Movement        gain = -7
subBus[ 8] VO: player, radio             gain = -8
subBus[ 9] VO: Event                     gain = -4
subBus[10] VO: In game/scripted, battle  gain = 2
subBus[11] Mix Group 1                   gain = -6
subBus[12] Mix Group 2                   gain = -6

subBus[1] Player Weapon sits at −11 dB. This is the most direct knob for "weapons louder," and it is a single number in a text file.

5. Cvars

TNO, read from WolfNewOrder_x64.exe (Finding: the strings are inside the EXE; that they are all registered as Cvars is Speculation, supported by the fact that other names from the same list function in-game verifiably):

s_volume_dB           s_voiceVolume_dB      s_muteMusic          s_noSound
s_reverb              s_showReverb          s_useOcclusion       s_occlusionDistance
s_lowPassFilter       s_useDistanceLowPass  s_maxSoundDistanceClamp
s_hdrFalloff          s_skipHDR             s_showHDR            s_showDucks
s_showSounds          s_drawSounds          s_showSoundsFilter   s_soundGroupFilter
s_maxEmitterChannels  s_debugMaxPlay        s_finiteSpeedOfSound s_tinnitusStart

s_volume_dB and s_voiceVolume_dB are accessible via the existing Start Argument path (+s_volume_dB 6), completely without an archive patch. Not verified in-game.

TNC, from the old scraped candidate list (since replaced by the registry catalogue, TNC-CVar-Catalog). Checked against the engine registry: s_printLoadedBanks, s_printMapBankReport and s_loadEditorBanks are not Cvars; the other nine are registered, and s_volume, s_volume_music, s_volume_vo and s_overridevolume are exposed to the retail console:

s_volume   s_volume_music   s_volume_vo   s_overridevolume   s_muteSound
s_muteEnvironments   s_reverbBus   s_skipOcclusion   s_printLoadedBanks
s_printMapBankReport   s_initForcePreLoadSoundBanks   s_loadEditorBanks

(The numerous s_bossshoot*, s_dogchase*, etc., belong to the Wolfenstein-3D engine within the game, not the mixer layer.)

6. What is changeable and what is not

Changeable with existing machinery

sound-, soundmixing-, impactSounds-, voiceover-, and footstepEvents-Decls are ordinary Decl Archive entries without a corresponding .bin counterpart. The existing patch path (wolfsdk/patch.py, deflate_exact, Backup + Journal) applies unchanged, including Revert.

This makes achievable:

• Volume globally (inheritences/master) or per family
• Mixbus Gains (soundmixing:default)
• Range, reverb, lowpass filter, priority, voice count
• Random selection: more/less variants, redirect to other existing samples — this is the practical way to achieve "other weapon sounds"
• Music assignment: music/*-Decls can point to other existing music tracks

Hypothesis, not verified in-game: that modified Sound Decls actually take effect. Structurally, everything points this way (same entry residue as the patchable entityDef/weapon), but it has not been verified.

Pitfall: 557 Decls carry an rmsTable for a pre-calculated loudness curve. If volume is increased without adjusting the curve, the HDR audio layer may compensate. Hypothesis. To bypass this, change the gain in soundmixing, which sits behind the HDR stage.

Changeable but without backup — replacing real audio data

Technically, this is the simplest case of the entire project: write an Ogg Vorbis stream to a known offset and adjust two BE-u32 fields (length @24 and @48) plus the granule in the 74-byte bsnf. The descriptor is a normal Archive entry, so it fits into the existing patch path. The slot offers median 8.6 KB of headroom, but for some sounds 0 bytes — a longer stream would overwrite the next sound there.

What argues against this, and significantly:

1. streamed.resources and english.streamed are not in master.index. Installation.ensure_backup() backs up exclusively the 14 chunkN.index files and records the lengths of chunkN.resources. The two streaming containers total 1.5 GB and are never backed up. A failure would only be fixable via Steam's file verification.
2. There is no Vorbis encoder in the standard library. New audio must be generated externally — the SDK is fixed to "standard library only".

Recommendation: do not touch this unless the containers are included in the backup. If so, first run ensure_backup to expand the two files and write exclusively shorter streams to keep offset and slot untouched.

Not changeable

• TNC Audio. The IDCL container format has been read (wolfsdk/idcl.py, verified against all 24 Retail .resources), but the unpacker is not connected and there is no write path. The audio data additionally resides in Wwise Banks for which no parser exists — reading the container does nothing about this. (Superseded 2026-10-04: banks are parsed and a replacement path is built, enabled in the Studio since 0.3.0; sections 9–11.)
• voicetrack blocks: 310 bytes per language, content undeciphered.
• .samplepack (Magic 1KPS, 253 items): variant bundles with their own streaming offsets in the entry residue. Not further investigated.

7. Part 2 — Wolfenstein II: The New Colossus

Just a status report. The IDCL container format has been read since wolfsdk/idcl.py (Head, Regions, String table, verified against all 24 Retail .resources); for the .pack files here it is still without consequence because the content is Wwise Banks and the unpacker is not connected.

What lies in the filesystem

base/sound/soundbanks/pc/ sound.pack 1,722,766,110 B IDCL english(us).pack 758,900,091 B IDCL patch_1_sound.pack 110,345,075 B IDCL patch_3_sound.pack 193,042,687 B IDCL patch_2_sound.pack 47,419,238 B IDCL patch_1_english(us).pack 18,712,107 B IDCL patch_3_english(us).pack 16,697,329 B IDCL patch_2_english(us).pack 12,209,825 B IDCL patch_4_sound.pack 132 B IDCL (empty) patch_4_english(us).pack 132 B IDCL (empty) wwise_project.bin 4,493,695 B "WPCv05r" prodspec.json 163,237 B Klartext-JSON

For each DLC (dlc/dlc_1|2|3/base/sound/soundbanks/pc/), the same set applies, including all languages: french(france), italian, russian, portuguese(brazil), spanish(mexico), spanish(spain). In total, 46 files: 34 .pack files, 4 .json files, 4 .bin files, totaling 3.15 GB.

The only extensions found throughout the game directory are: bk2 (122, Bink videos), pages (42, virtual textures), cache (39), pack (34), texdb (27), resources (27), cfg (12), json (11), bin (6).

There are no .streamed files and no .bsnd assets. The audio layer has been fully migrated to Wwise. This is a finding from the filesystem, not an inference: wwise_project.bin carries the magic number WPCv05r and lists in plain text 1,394 .bnk and 78,558 .wem names — .wem is the Wwise Encoded Media Format.

The Plain-Text Configuration

prodspec.json is unencrypted and lists every single soundbank:

{ "banksize": 1699459, "name": "sb_mus_combat_suites.bnk", "streamsize": 102838598 }

• 937 Banks, patchlevel: 4
• Sum of banksize: 1.02 GB, Sum of streamsize: 8.45 GB
• Bank families: sb_amb (55 Ambience), 30 sb_vo_* per language (English(US), French(France), Italian, Spanish(Spain), Russian, Spanish(Mexico), Portuguese(Brazil)), sb_pr 24, sb_events 23, sb_mus 20, sb_ch 18, sb_preset 16, sb_pl 13, sb_ai 7, sb_fx 5
• Largest streams: sb_mus_combat_suites.bnk (102.8 MB stream), <Language>/sb_vo_ai.bnk (~70.9 MB each)
• The maps key maps bank lists to game cards — hence also the Cvars s_printMapBankReport and s_initForcePreLoadSoundBanks.

wwise_project.bin also lists languages that are not installed (German/sb_vo_c2.bnk, Japanese/sb_vo_c2.bnk) — it appears to be the complete project directory, not the release list. Hypothesis.

What This Means Practically

In TNC, for now, only the Cvar path works: s_volume, s_volume_music, s_volume_vo, s_overridevolume. The IDCL container is read, but mods=False — there is no write path for Part 2 — and the mixing itself resides in the Wwise banks for which no parser exists. (Superseded 2026-10-04: the packs are plain IDCL archives that wolfsdk/wwise.py reads, banks included; a sound replacement is built and enabled in the Studio since 0.3.0, sections 9–11.)

8. Reproduction

python tools/probe_audio.py

Runs for approximately 70 seconds, reads 203,843 index entries, and checks all 61,318 sample descriptors against the Ogg sidechains in the streaming containers. Writes nothing.

9. TNC Sound Replacement: How a WAV Becomes a Game Sound

Built in WolfSDK Studio; held back in 0.2.0, enabled since 0.3.0, not yet tried in game (section 11).

• Input: a 16-bit PCM WAV with the game sound's channel count and at most its length (0.01 s slack for export rounding), any sample rate. It stays PCM, the game's other codec: there is no Vorbis encoder here.
• Output: a PCM WEM laid out like the game's own PCM sounds (fmt tag 0xFFFE, data at offset 64), saved as asset wem:<sound id>.
• Writing: the loader rewrites every sound pack holding the sound, in every patch level. A loose .wem is replaced whole; in a bank the new slice goes to the end of DATA (16-aligned), its DIDX row is repointed, and the old slice stays unreferenced.
• Every Sound object (HIRC type 2) of that medium is switched to the PCM plugin 0x00010001, with the size field of its stream type. Measured: the plugin field names the codec: 0x00040001 in all 129,705 Vorbis sources, 0x00010001 in all 409 PCM ones.

Keywords: replace sound, WAV, WEM, Wwise, sound mod, Sound ersetzen, Geräusch austauschen

10. TNC Sounds That Cannot Be Replaced Yet

The Studio refuses these with a reason (wolfsdk/wwise.py):

Keywords: music, prefetched, streamed sound, cannot replace, not supported, Musik ersetzen, geht nicht

11. Why Sound and Video Replacement Was Held Back in WolfSDK 0.2.0

Enabled since 0.3.0. In release 0.2.0 the Studio refused sound and video replacements: Revert did not restore the replaced sound packs and video files (the check tools/verify_studio_media.py, "after revert every file of the tree is md5-identical", failed). Since 0.3.0 Revert restores them and the Studio replaces sounds and videos for The New Colossus. Not yet tried in game.

The video replacement is built the same way and was held back with it: a Bink 2 file made with RAD's tools, with the same size and frame rate as the game's video and at most its frame count. The Studio checks it and opens it with the game's own bink2w64.dll; it does not convert. The loose file is replaced whole (asset video:<path>).

Neither path is confirmed in game.

Keywords: held back in 0.2.0, sound replace, video replace, Bink, cutscene video, revert, zurückgehalten, Video ersetzen