A creative multitrack DAW for Windows 64-bit

面向 Windows 64 位的创意多轨数字音频工作站

HarmonicDAW is a multitrack audio + MIDI workstation that puts musical intuition into the architecture itself. The same surface any experienced DAW user expects — tracks, regions, automation, plugin chains, master bus — but with a top rail of creative effects (Glitch, Shiftermate, the harmonic Pitch slider) always within reach, a set of built-in instruments and processors that cover most starting needs without external plugins, and a per-channel "effect group" that lets every track sport its own pitched chunk repeater, glitch chopper, fractal pitch sequencer and EQ before or after its plugin chain. All 64-bit internal precision, all undoable, all stored in a single human-readable .psy project file.

HarmonicDAW 是一款多轨音频 + MIDI 工作站,将音乐直觉融入软件架构本身。 任何有经验的 DAW 用户都熟悉的界面——轨道、区域、自动化、插件链、主总线 ——加上始终触手可及的顶部创意效果栏(Glitch、Shiftermate、Pitch 滑块), 一套无需第三方插件即可满足大多数起步需求的内置乐器与处理器,以及每个声道独有的"效果组"—— 让每条轨道在其插件链前后都可拥有自己的音高块重复器、Glitch 切片器、分形音高音序器和 EQ。 全程 64 位内部精度,完全可撤销,所有内容保存在单个人类可读的 .psy 工程文件中。

The Harmonic Grid degree-locked audio stacks

Harmonic Grid 按音程度数锁定的音频堆叠

Drop an audio region onto a track, switch to Harmonic mode, pick a chromatic degree from the menu — degree 0 (tonic) through degree 11, with degree 5 (subdominant) and degree 7 (dominant) labelled as familiar landmarks. Keyboard shortcut Shift+0–9 (and Shift+_ / Shift++) snap a selected region straight to that degree. The audio instantly plays tuned to that interval, in time with everything else.

将一个音频区域拖到轨道上,切换为 Harmonic 模式,在菜单中选择一个半音度数 ——从 度数 0(主音) 到 度数 11, 其中 度数 5(下属音) 和 度数 7(属音) 已标注为熟悉的参照点。键盘快捷键 Shift+0–9(及 Shift+_ / Shift++) 可将选中区域直接定位到对应度数。音频会立即按该音程调音,并与其余内容保持同步。

Stack as many duplicates as you want; each sits at its own chromatic degree, all locked to the project tonic. Vocal harmonies, pitched percussion, drone ladders, evolving overtone series — all from a single source clip, no recutting, no external pitch-shifter plugin, no time drift.

想堆多少副本就堆多少;每一份占据自己的半音度数,全部锁定到工程主音。 人声和声、有调打击、嗡鸣音梯、演变的泛音列——全部源自同一段素材,无需重切片、 无需外部移调插件、无时间漂移。

Region Macros harmonic transpose, random chop & replace

区域宏操作 谐波移调、随机切片与随机替换

Right-click any audio region for a context menu packed with harmonic and creative operations beyond the usual cut / copy / paste:

右键任意音频区域,打开包含一系列谐波与创意操作的上下文菜单,远不止常规的剪切 / 复制 / 粘贴:

The same context menu also provides Cut at Transients (note-onset detection), Gain Bend, Macro Cascade, lock/mute/colour per region, and fast tool switching (Cursor / Scissors / Glue / Mute) — all one right-click away.

同一上下文菜单还提供 Cut at Transients(音符起音点检测)、 Gain Bend、Macro Cascade、每区域独立锁定 / 静音 / 颜色, 以及快速工具切换(光标 / 剪刀 / 粘合 / 静音)——全部只需右键一次。

AI Generation instruments, vocals & continuation — 100% local

AI 生成 乐器、人声与续写 —— 100% 本地化运行

A full local AI generation stack runs entirely on your own machine — nothing is sent to a server, no subscriptions, no per-generation fees. Once a model is downloaded it works offline.

完整的本地 AI 生成栈，完全运行在你自己的电脑上——不向任何服务器发送数据,无订阅,无按次生成费用。 大模型下载一次后即可离线使用。

The install is lightweight: three independent installers — Install_Audio.bat, Install_SA3.bat, Install_Vocals.bat — each set up an isolated environment via uv (no Python setup, no accounts). Engines run as localhost-only servers (ports 8765 / 8766 / 8767) that never leave your computer. AI is generative — not every take is perfect; reroll the seed, refine the prompt, keep the best.

轻量化安装:三个独立的安装器——Install_Audio.bat、 Install_SA3.bat、Install_Vocals.bat—— 各自通过 uv 建立隔离的运行环境(无需自行安装 Python,无需账号)。 各引擎作为仅限本机(localhost)的服务运行(端口 8765 / 8766 / 8767),数据从不离开你的电脑。 AI 是生成式的——并非每一次都完美;重新调整种子、优化提示词,留下最好的那一部分。

Automation, Tempo & Reverse five curves, signed time

自动化、速度与反向播放 五种曲线,带符号时间

Every track parameter — volume, pan, the per-channel Pitch effect, every plugin knob — carries an automation lane. Curves between points come in five shapes: Linear, Exponential, Logarithmic, S-curve, and Stepped for bar-locked toggles. Read / Write / Pull modes per lane; a track-wide write-arm lets a whole knob sweep get captured at once.

每条音轨的全部参数——音量、声像、每通道独立的 Pitch 效果、以及所有插件旋钮——均支持专用自动化轨道。曲线点之间的过渡提供五种形状:Linear(线性)、Exponential(指数)、Logarithmic(对数)、S-curve(S 曲线) 以及 Stepped(阶梯)(适用于小节锁定的开关式切换)。每条自动化轨道均支持 Read / Write / Pull 三种模式;轨道级 Write Arm 功能可一次性捕获整个旋钮的完整扫动动作。

The project's Tempo Track is automation too — the BPM is a first-class lane that you can ramp, step, or hold. Combined with the Reverse Playback transport mode, you can play a project backward at variable tempo and watch every regions's waveform render in reverse with FX bypassed for clean playback.

项目的 Tempo Track(速度轨) 本身也是自动化轨道——BPM 作为一类核心自动化曲线,可进行连续渐变(ramp)、阶梯跳变(step)或保持(hold)。结合 Reverse Playback(反向播放) 传输模式,您可以让整个工程以可变速度反向播放,同时观察所有区域的波形实时反向渲染,并自动旁通 FX 处理以确保纯净回放。

The Snap / Grid system has a Rhythmic mode (1/64 → 1/1, triplets and dotted included) and a Harmonic mode (frequency-based for the new HarmonicBufferPitch). Loop markers, region snap, and the playhead all respect whichever mode is active.

吸附/栅格系统 提供两种模式:

• Rhythmic 模式(节奏模式):支持 1/64 至 1/1 以及三连音、附点等传统节奏划分;
• Harmonic 模式(谐波模式):基于频率设计,专为全新的 HarmonicBufferPitch 功能优化。

循环标记、区域吸附与播放头均会严格遵循当前激活的模式。

Multi-Loop four independent loop markers, Shift+L/U/V/W

多循环 四个独立循环标记，Shift+L/U/V/W

Set up to four independent loop markers — LOOP 1 (Shift+L), LOOP 2 (Shift+U), LOOP 3 (Shift+V), LOOP 4 (Shift+W) — each shown on the timeline in its own colour (amber / cyan / magenta / green) with a small label. When the Loop switch is on and more than one loop is set, the transport cycles through them in order — 1 → 2 → 3 → 4 → 1 → … — and keeps repeating until you stop.

最多可设置四个独立的循环标记——LOOP 1(Shift+L)、 LOOP 2(Shift+U)、LOOP 3(Shift+V)、 LOOP 4(Shift+W)——每个标记在时间线上以独有颜色 (琥珀色 / 青色 / 品红色 / 绿色)和文字标签显示。开启循环开关且设置了多个循环时, 传输将按顺序循环切换——1 → 2 → 3 → 4 → 1 → ……直到手动停止。

• Create from any selection — lower loops auto-fill consecutively before the new one, same length.
• Drag the coloured handles to resize any loop; it stays beat-anchored through tempo changes.
• Right-click the loop strip to clear individual loops or all at once.
• Export with the Loop switch on and all active loops render back-to-back into one continuous WAV — the perfect live-deck bounce.
• If the Loop switch is off, the markers stay visible as coloured labels without affecting playback.

• 从任意选区创建——低编号循环自动在新循环前方以相同长度依次排列。
• 拖动彩色手柄可调整任意循环的长度;循环标记以节拍为锚点,在速度变化后保持网格对齐。
• 右键循环条可单独清除某个循环或一键清除全部。
• 开启循环开关后执行 导出,所有激活的循环将依次渲染为一个连续 WAV 文件——完美的现场演出素材导出。
• 若循环开关关闭,标记仍以彩色标签形式保持可见,不影响播放。

Built-in Instruments no third-party plugins required

内置乐器 无需第三方插件

Kick

Drum synthesizer with editable pitch envelope

底鼓合成器,可编辑音高包络

A pitch-and-amplitude shaper for kick drums. Drag points on the pitch curve from sub-50 Hz tails up through 3+ kHz click transients; mirror the same control on the amp envelope. Length, gain, multi-segment shapes — every classic 808 to industrial thump in one editor.

一款专为底鼓打造的音高与幅度塑形器。在音高曲线上拖动节点,从 50 Hz 以下的低频尾音 一直延伸到 3 kHz 以上的咔嗒瞬态;音量包络上拥有相同的操作。长度、增益、多段形状 ——经典 808 到工业风重击,统统在同一个编辑器中完成。

Bass

Subtractive synthesizer

减法合成器

Two oscillators (saw / square / triangle / pulse with detune, voices, octave, phase), a hard or soft low-pass filter, an amp envelope, two assignable mod envelopes, two LFOs. Routes mod signals to cutoff, pitch, pan, volume, or any of the synth's own parameters. Single-instrument basslines, leads, arps.

两个振荡器(锯齿 / 方波 / 三角 / 脉冲,带失谐、复音、八度、相位), 硬 / 软低通滤波器,放大器包络,两条可指派的调制包络,两个 LFO。 调制信号可路由到截止频率、音高、声像、音量,或合成器自身的任何参数。 一台乐器搞定贝斯线、主奏、琶音。

Sparx

Sound effect synthesizer

音效合成器

Twelve preset categories — pickup, laser, explosion, power-up, hit, jump, blip, pluck, beep, heartbeat, plus init and random. Click a category to roll a fresh sound; Mutate to evolve it; Follow MIDI Pitch to tune it to incoming notes. Modern descendant of the bfxr / sfxr family.

十二种预设分类——拾取、激光、爆炸、加电、击中、跳跃、闪烁、拨弦、蜂鸣、心跳, 外加初始和随机。点击分类即可生成一段新音效;Mutate 让它继续进化; Follow MIDI Pitch 跟随传入的音符调音。bfxr / sfxr 家族的现代后裔。

SAM

Software Automatic Mouth — speech synthesizer

Software Automatic Mouth——语音合成器

Type a word, send a MIDI note, and the original 1982 SAM speech engine reads it back in robot voice — Speed, Pitch, Mouth, Throat, Volume controls plus a Singing toggle that slaves the pitch to MIDI. Iconic, ridiculous, surprisingly musical when stacked with reverb.

敲入一个单词,发送一个 MIDI 音符,1982 年原版 SAM 语音引擎就会以机器人嗓音念出它 ——Speed、Pitch、Mouth、Throat、Volume 控件加一个 Singing 开关, 让音高跟随 MIDI。标志性、荒诞,叠上混响后却出奇地有音乐性。

Psyko

Dual-oscillator wavetable synthesizer

双振荡器波表合成器

Two wavetable oscillators that load real exported wavetables — each with its own table, position and unison. OSC B phase-modulates OSC A (FM-from-B) for metallic, vocal, glassy timbres a single table can't reach. Draw-your-own breakpoint LFO routed to a 24 dB morphing filter, tempo-synced, with a 3D crystalline wavetable display that morphs as you sweep the position. Edit, save, and reset tables in place.

两个可加载真实导出波表的波表振荡器——各自拥有独立的波表、位置与齐奏。 OSC B 对 OSC A 进行相位调制(FM-from-B),带来单一波表无法企及的金属、人声、 玻璃质感音色。可自绘断点 LFO,路由到 24 dB 形变滤波器并与速度同步, 配以随位置扫描而形变的 3D 水晶波表显示。波表可就地编辑、保存与重置。

Supersaw

16-voice unison saw lead

16 音齐奏锯齿主奏

Sixteen PolyBLEP saws in unison — the festival / hardstyle lead wall. Wide detune and stereo spread, a sub-octave layer, ladder / SVF / diode 24 dB filters with a filter envelope, key-tracked high-pass, drive, vibrato, and a true-peak limiter so the stack stays huge without spitting. New SHAPE and HUMANIZE knobs give every unison voice its own band-limited waveform width and re-roll it on each note — spreading the spectrum and adding analog-style drift so the lead never sounds static.

十六个 PolyBLEP 锯齿齐奏——节日 / Hardstyle 风格的主奏声墙。宽幅失谐与立体声扩展、 低八度叠层、ladder / SVF / diode 三种 24 dB 滤波器及滤波包络、跟键高通、过载、颤音, 外加真峰值限制器,让声音庞大却不破音。新增的 SHAPE 与 HUMANIZE 旋钮为每个齐奏声部 赋予各自的带限波形宽度,并在每次触发时重新随机——拓宽频谱、加入模拟风格的漂移, 让主奏永不呆板。

Cluster

Generative sine-cluster drone

生成式正弦簇持续音

Hold a chord and Cluster grows a shifting constellation of pure sine partials that drift and beat against each other — a living drone bed, not a static pad. The played note sets the cluster pitch; SEED fixes the voicing; partial count, frequency range, LFO rate and depth, voicing, spread and a long rise / fall swell shape how it breathes. Its own supersampled backdrop animates the cluster as it plays — built for intros, breakdowns and cosmic floors.

按住一个和弦,Cluster 便生长出一片不断变幻的纯正弦分音星群,彼此漂移、拍频—— 是会呼吸的持续音床,而非静止的铺底。所弹音符决定簇的音高;SEED 固定声部排布; 分音数量、频率范围、LFO 速率与深度、声部、立体声扩展,以及悠长的渐入 / 渐出涌动, 共同塑造它的呼吸。专属的超采样背景会随演奏让整片簇群动起来——为前奏、过渡段 与宇宙感铺底而生。

GrainMate

Granular sampler

颗粒采样器

Drop in any audio file and GrainMate scatters it into a cloud of grains you play polyphonically across the keyboard. A scanning read-head crawls the sample while the grain engine rebuilds it in real time — position, scan, grain size, overlap, spray, pitch, jitter and spread, with the scan lockable to project tempo. Flip Free-run and it becomes a hands-off, self-scanning granular drone with no note required.

拖入任意音频文件,GrainMate 便将其打散成一片颗粒云,可在键盘上复音演奏。 扫描读头缓缓爬过样本,颗粒引擎实时重建声音——位置、扫描、颗粒大小、叠加、喷散、 音高、抖动与立体声扩展,扫描速度还可锁定到工程速度。开启 Free-run, 它即化身无需按键、自行扫描的颗粒持续音。

Kaeru

AI singing voice instrument — Kaerlighed voice

AI 演唱声音乐器——Kaerlighed 人声

Type lyrics into MIDI notes, draw a melody, hit Re-render all. Kaeru sings it in Kaerlighed's real voice — trained on studio recordings using a DiffSinger diffusion model with Whisper word-level phoneme alignment. Vibrato, Portamento, Dynamics and Register sliders. Runs 100% locally via a lightweight sidecar; no cloud, no subscription, no latency cap.

在 MIDI 音符中输入歌词,绘制旋律,点击 Re-render all。 Kaeru 将以 Kaerlighed 的真实嗓音 演唱——基于工作室录音训练, 采用 DiffSinger 扩散模型与 Whisper 词级音素对齐技术。 提供颤音、滑音、动态与音区滑块。通过轻量级本地进程 100% 本地运行, 无需云端,无订阅,无速率限制。

Kaeru synthetic singing voice — Kaerlighed 1.0

Kaeru 合成歌唱声 —— Kaerlighed 1.0

Kaeru is a MIDI instrument that sings your lyrics in a real trained voice. Write notes in the piano roll, type a word into each note, set pitches — Kaeru renders the phrase through a local DiffSinger acoustic model trained on studio recordings of vocalist Kaerlighed. No cloud, no per-render fee, no generation queue. It runs as a lightweight localhost server that the DAW queries silently in the background.

Kaeru 是一款用真实训练声音演唱歌词的 MIDI 乐器。 在钢琴卷帘中绘制音符,为每个音符输入一个词,设置音高—— Kaeru 通过本地 DiffSinger 声学模型渲染短语,该模型基于歌手 Kaerlighed 的工作室录音训练。无需云端,无按次渲染收费,无生成队列。 以轻量级 localhost 服务器形式运行,DAW 在后台静默查询。

• Voice Picker — switch between installed voices; Kaeru 1.0 (Kaerlighed) ships in v1.9.1.
• Vibrato — pitch vibrato depth on sustained notes (default 0.15).
• Portamento — pitch glide between notes (default 0.15).
• Dynamics — overall vocal intensity from breathy to full power (default 1.0).
• Register — shift voice up/down ±24 semitones; write in any octave, hear it correctly (default 0 st).
• Re-render all — one click re-renders all Kaeru regions; only changed regions re-process.
• Double-click any slider to reset to default.

• Voice Picker（声音选择）——在已安装的声音之间切换；v1.9.1 随附 Kaeru 1.0（Kaerlighed）。
• Vibrato（颤音）——持续音符上的音高颤音深度（默认 0.15）。
• Portamento（滑音）——音符间的音高滑动（默认 0.15）。
• Dynamics（动态）——从气声到全力的整体人声强度（默认 1.0）。
• Register（音区）——将声音上下移动 ±24 个半音；在任意八度中写谱，听起来都正确（默认 0 st）。
• Re-render all（全部重新渲染）——一键重新渲染所有 Kaeru 区域；仅重新处理已更改的区域。
• 双击任意滑块可重置为默认值。

What's New in 1.9.5 effects rebuilt, CLAP hosting, new processors

1.9.5 新增内容 效果器重制、CLAP 宿主、新处理器

• Reverb rebuilt — 32-line FDN sized to Schroeder's mode-density criterion, energy-normalised injection, permanent DC blocker, mono-safe stereo, and a new M/S DECAY that lets the side ring up to 2:1 longer than the centre.
• Delay — glided, Hermite-interpolated read pointer (no more clicks on division changes), a tap ladder extended to 1/512, single MIX knob, live echo display.
• Force Field Delay — automation now works in both directions on every control including the menus; SYNC's gating of TIME is visible instead of silent; double-click resets any knob to its default.
• Finalizer — Spectral mode removed, UPWARD compression added.
• Modal — new parallel modal resonator bank, up to 512 modes.
• VST2 and CLAP hosting — CLAP implemented directly against the C API.
• Compose from Folder — build arrangements from a folder of samples.
• Slice Decimate — new force-field methods for the frame-drop retimer.
• Interface — bloom off by default and genuinely sharper geometry across the application; square Compressor with a live transfer-characteristic display; one shared theme across the modulation, Compose and Pitch Shift dialogs.

• 混响重制——按 Schroeder 模态密度准则设计的 32 线 FDN、 能量归一化输入注入、常驻直流阻断、单声道安全的立体声， 以及新增的 M/S DECAY（侧向可比中心长达 2:1）。
• 延时——滑移式 Hermite 插值读指针（切换分割不再有咔哒声）、 阶梯扩展至 1/512、单一 MIX 旋钮、实时回声显示。
• Force Field Delay——包括下拉菜单在内的所有控件现已双向支持自动化； SYNC 对 TIME 的屏蔽状态可见而非静默；双击任意旋钮恢复默认值。
• Finalizer——移除 Spectral 模式，新增 UPWARD 向上压缩。
• Modal——全新并联模态谐振器组，最多 512 个模态。
• VST2 与 CLAP 宿主——CLAP 直接基于 C API 实现。
• Compose from Folder——从一个采样文件夹生成编排。
• Slice Decimate——为掉帧重定时器新增力场方法。
• 界面——全应用默认关闭辉光并采用真正更锐利的几何； 方形 Compressor 配实时传输特性曲线显示；调制、Compose 与 Pitch Shift 对话框统一为同一主题。

Built-in Effects twenty-two processors, 64-bit internal

内置效果器 二十二种处理器，64 位内部精度

Bitcrusher

Sample-rate divider + bit-depth crusher

采样率分频 + 位深压碎

Two independent degradations running together. A sample-rate divider (×1 to ×64) holds each input sample for N successive samples before emitting a new one — the classic stair-stepping that folds aliasing back down into the audible band rather than filtering it away, which is what gives digital crush its metallic edge. Alongside it, a bit-depth quantiser from 24 bits down to 1, re-rounding every sample to a coarser grid so quiet passages break up into steps while loud ones stay comparatively intact. The two interact: heavy division with high bit depth gives ringing aliasing, light division with low depth gives gritty quantisation noise, and both together give console-era destruction. Mix and output trim let you blend the damage back under the source rather than replacing it, so it works as a parallel texture as readily as an effect.

两种独立的劣化同时运行。采样率分频器（×1 至 ×64） 将每个输入采样保持 N 个连续采样后才输出新值—— 这种典型的阶梯效应会把混叠折返回可听频段，而不是将其滤除， 这正是数字破坏特有的金属质感来源。与之并行的是 位深量化器，可从 24 位一路降到 1 位， 把每个采样重新舍入到更粗的网格上，因此安静段落会碎裂成台阶， 而响亮段落相对完好。两者相互作用：高分频加高位深会产生嗡鸣式混叠， 低分频加低位深则得到粗砺的量化噪声，两者同时推满则是游戏机时代的彻底破坏。 Mix 与输出微调可让你把这些损伤混回原信号之下而非取代它， 因此它既能作为效果器，也同样适合作为并行纹理层使用。

Reverb

Algorithmic reverb

算法混响

Rebuilt in v1.9.5. A 32-line Feedback Delay Network with Hadamard mixing, sized to Schroeder's mode-density criterion — strong modes moved from 82 Hz apart to 10 Hz, which is the difference between ringing and diffusing. Energy-normalised injection means the tail gets longer, not louder, as REVERB TIME rises. Permanent DC blocker in every line. Equal-power panning with a shared per-line sign: nothing anti-phase, every line survives a mono sum. New M/S DECAY lets the side ring up to 2:1 longer than the centre — depth without smearing the middle of your mix — plus mono bass below 160 Hz. DIFFUSE and RESONANT modes, EFFICIENCY trading 32 / 16 / 8 lines.

v1.9.5 重制。采用 Hadamard 混合的 32 线反馈延迟网络（FDN）， 按 Schroeder 模态密度准则设计——强模态间距从 82 Hz 缩小到 10 Hz， 这正是"嗡鸣"与"扩散"之间的区别。能量归一化的输入注入意味着提高 REVERB TIME 时 尾音变长而非变响。每条延迟线都有常驻直流阻断器。等功率声像配合每线共享符号： 无反相内容，所有线在单声道求和后都保留。新增 M/S DECAY，可让侧向能量比中心 长达 2:1 ——获得深度而不糊掉混音中心——并在 160 Hz 以下实现单声道低频。 DIFFUSE 与 RESONANT 两种模式，EFFICIENCY 可在 32 / 16 / 8 线之间权衡。

Delay

Ping-pong stereo cross delay

乒乓立体声交叉延时

Two cross-fed delay lines so echoes alternate L → R → L → R. v1.9.5 replaces the integer read pointer with a 60 ms glide and 4-point Hermite interpolation, so changing division is a tape-style pitch slide instead of a click. The tap ladder now runs to 1/512 — twelve note values on one knob handing over to a 37-step harmonic-degree ladder on the next, so reading left to right the delay only ever gets shorter. At 120 BPM the top of the ladder is 3.9 ms (256 Hz), well into pitched Karplus-Strong territory. Single MIX knob, all-knob panel, and a live display of the actual echo pattern — tap spacing, feedback decay, damping and ping-pong.

两条交叉反馈的延时线，使回声在 L → R → L → R 之间交替。v1.9.5 将整数读指针改为 60 毫秒滑移加 4 点 Hermite 插值，因此切换分割时是磁带式的音高滑动 而非咔哒声。分割阶梯现已延伸至 1/512——一个旋钮上十二种音符时值， 交接到下一个旋钮上 37 级的谐波度数阶梯，从左向右读取时延时只会越来越短。 在 120 BPM 下阶梯顶端为 3.9 毫秒（256 Hz），已进入有音高的 Karplus-Strong 领域。 单一 MIX 旋钮，全旋钮面板，并实时显示真实的回声图案——间距、反馈衰减、阻尼与乒乓。

Compressor

Feed-forward dynamic range control

前馈式动态范围控制

A feed-forward design with all the standard controls — threshold, ratio, attack, release, knee, make-up gain and mix — computing gain reduction from the input rather than the output, so the response is predictable and the knee behaves the way the curve says it will. The soft knee interpolates smoothly through the threshold region instead of switching abruptly, which is the difference between compression you hear working and compression you only notice when it stops. Parallel MIX means you can keep transients intact underneath a heavily compressed layer without a second bus. The panel is built around a live transfer characteristic — input dB across, output up, a unity diagonal for reference, and the actual curve computed from your current settings including the knee interpolation. Knee width is something you can see rather than infer. A dot rides the curve at the current operating point, and a horizontal GR meter is locked to the display's width beneath it.

前馈式设计，配备全部标准控件——门限、压缩比、启动、释放、拐点、补偿增益与混合—— 增益衰减由输入信号而非输出信号计算，因此响应可预测， 拐点的表现与曲线所示完全一致。软拐点在门限区域平滑过渡而非突然切换， 这正是"听得出在工作的压缩"与"只有停止时才会察觉的压缩"之间的区别。 并行 MIX 让你无需额外总线，就能在重度压缩层之下保留完整瞬态。 面板围绕实时传输特性曲线构建——横轴为输入 dB、纵轴为输出， 以单位对角线作参照，并根据当前设置（含拐点插值）实时计算真实曲线。 拐点宽度是可以看见的，而不需要凭空推断。曲线上有一个光点随当前工作点移动， 下方是与显示区等宽的横向增益衰减表。

Opto Comp

LA-3A-style optical compression

LA-3A 风格光学压缩

An optical-cell compressor in the LA-3A lineage. Where a feed-forward compressor computes exactly the gain reduction its settings specify, an opto cell is a physical component with its own behaviour: the release is programme-dependent, fast at first and then slowing as the cell recovers, and the attack softens as gain reduction deepens. You do not set that curve — it emerges, which is why opto compression sits so naturally on vocals and bass where a linear release would pump. The result flatters sustained material and largely ignores fast transients, so it levels a performance rather than shaping individual hits. PEAK REDUCTION drives the cell, GAIN makes up the loss, and MIX blends the compressed signal back against the source for parallel work. The panel reads as a metering instrument — crisp cyan detector controls with amber on the one knob that adds level.

属于 LA-3A 传承的光学单元压缩器。前馈压缩器会精确计算设置所指定的增益衰减， 而光学单元是一个具有自身物理行为的元件：其释放是随节目素材变化的—— 起初较快，随着单元恢复而逐渐变慢；启动时间也会随着衰减加深而变柔。 这条曲线不是你设定的，而是自然涌现的，这正是光压在人声与贝斯上如此自然的原因， 换成线性释放就会产生"喘息"。其结果是美化持续性素材、基本忽略快速瞬态， 因此它是在为一段演奏找平，而不是雕刻单个音头。 PEAK REDUCTION 驱动光学单元，GAIN 补回损失的电平， MIX 则将压缩信号与原始信号混合以实现并行压缩。 面板呈现为一台计量仪器——锐利的青色检测控件，唯一提升电平的旋钮采用琥珀色。

TranceGate

Tempo-synced 4-slot gate

速度同步 4 槽门

A tempo-synced amplitude gate that chops audio on a rhythmic grid. Each step is either ON (passes) or OFF (mutes), and an ADSR envelope shapes every ON step's edges — without that, switching a signal on and off produces a click on every transition, which is why a gate built from raw on/off states sounds like static rather than rhythm. Attack softens the opening, release the close, so you can go from a hard stutter to a breathing pulse with the same pattern. Four sequence slots chain together, so a pattern can evolve across four bars instead of repeating every one — the difference between a gate that drives a track and one you have to automate around. Step count, rate and grid division are all tempo-locked, so it stays in time through tempo changes.

与速度同步的振幅门，按节奏网格切碎音频。每一步要么 ON（通过）要么 OFF（静音）， 并由 ADSR 包络塑造每个 ON 步的边缘—— 没有它，信号的开关会在每次切换时产生咔哒声， 这正是仅用原始开关状态构建的门听起来像杂音而非节奏的原因。 Attack 柔化开启、Release 柔化关闭，因此同一个图案既可以是硬朗的断奏， 也可以是呼吸般的脉动。四个序列槽可串联， 因此图案可以跨四个小节演化而非每小节重复—— 这正是"能带动整首曲子的门"与"必须靠自动化补救的门"之间的区别。 步数、速率与网格分割全部锁定到速度，因此在变速时依然保持同步。

Finalizer

Mastering limiter

母带限制器

Three modes — Transparent (clean ceiling guard), Modern (gentle saturation), Crisp (aggressive, brighter). Threshold, ceiling, character, stereo link, transient emphasis, true-peak detection and soft clip. Live input / output / GR meters. New in v1.9.5: UPWARD. Every other tool here works on peaks; upward compression is the other half — it raises material below the threshold and leaves everything above it untouched, so average level rises with no gain reduction to pump. That is categorically cleaner than more limiting. Runs before the limiter, linked so the stereo image cannot wander on quiet passages, up to +10 dB of lift.

三种模式——Transparent（干净的天花板保护）、Modern（温和饱和）、 Crisp（激进、更明亮）。门限、天花板、特性、立体声联动、瞬态强调、 真峰值检测与软削波。实时输入 / 输出 / GR 表头。 v1.9.5 新增：UPWARD（向上压缩）。这里其他工具都作用于峰值； 向上压缩是另一半——它提升门限以下的素材，完全不触碰门限以上的部分， 因此平均电平上升而没有任何增益衰减可"喘息"。这在本质上比继续限制更干净。 置于限制器之前，左右联动以免安静段落的立体声像漂移，最高提供 +10 dB 提升。

Analyzer

1/3-octave spectrum analyzer

1/3 倍频程频谱分析仪

A pure pass-through spectrum analyser — it modifies nothing, so you can leave it anywhere in a chain without consequence. Thirty-one ISO 1/3-octave bands from 20 Hz to 20 kHz, which is the standard used for acoustic measurement rather than a linear FFT display: bands are spaced the way hearing is, so an even-looking readout actually corresponds to an even-sounding balance. A linear FFT gives most of its resolution to the top two octaves, where you need it least. Insert it before and after a processor to see exactly what that processor did, or park it on the master to check spectral balance against a reference while you mix.

纯直通式频谱分析仪——它不改变任何信号，因此可以放在插件链的任何位置而无副作用。 从 20 Hz 到 20 kHz 的三十一个 ISO 1/3 倍频程频段， 这是声学测量使用的标准，而非线性 FFT 显示： 频段的分布方式与人耳听觉一致，因此看起来平整的读数确实对应听起来平衡的声音。 线性 FFT 会把大部分分辨率分配给最高的两个八度，而那恰恰是最不需要的地方。 把它插在某个处理器前后即可精确看出该处理器做了什么， 或常驻在主总线上，在混音时对照参考曲目检查频谱平衡。

Cyborg Pitch

Spectral pitch shifter

频谱变调器

Two genuinely different engines rather than one algorithm stretched across both jobs. A phase-locked vocoder handles musical pitch shifting — it keeps the phase relationships between a partial and its neighbours intact, which is what stops shifted material smearing into the characteristic "underwater" artefact. Separately, a spectral bin-remapper produces robotic and inhuman character by moving energy between frequency bins with no attempt to preserve the harmonic relationships at all. Forcing one engine to do both is exactly what makes most pitch tools sound either clean but sterile or characterful but broken. Formant control is independent of pitch, so you can move a voice down two octaves while keeping the vocal tract where it was, or shift the formants alone to change apparent body without touching the note.

这是两套真正不同的引擎，而不是把一个算法硬拉去做两件事。 相位锁定声码器负责音乐性的移调—— 它保持分音与相邻分音之间的相位关系完好， 这正是移调后素材不会糊成典型"水下"人工痕迹的原因。 另一套独立的频谱 bin 重映射器则通过在频率 bin 之间搬移能量、 完全不试图保留谐波关系，来制造机械与非人的音色。 强迫单一引擎同时完成这两件事，恰恰是大多数移调工具要么干净但呆板、 要么有个性但破碎的原因。共振峰控制与音高相互独立， 因此你可以把人声降低两个八度而保持声道特征不变， 也可以只移动共振峰来改变听感上的"体积"而不触碰音高。

Spectral Freeze

Frame freeze + infinite sustain

频谱冻结 + 无限延音

Hit FREEZE and the current spectral frame is latched and sustained indefinitely — not a looped buffer, which repeats and reveals its seam, but a single analysed frame resynthesised continuously, so it holds forever without a loop point to hear. That is the distinction that makes it useful for drones: there is nothing periodic to give it away. The frozen frame can then be pitched independently of the source it came from, so one held chord becomes a whole pad by freezing it at several pitches. Because the freeze captures a moment rather than a duration, where you place it matters — freeze the attack of a note and you get a bright, edgy sustain; freeze the tail and you get something soft and hollow from the same source.

按下 FREEZE，当前的频谱帧就会被锁存并无限延续—— 它不是循环缓冲（循环会重复并暴露接缝），而是对单个已分析帧的持续重合成， 因此可以永远延续而听不到任何循环点。这正是它适合做持续音的关键： 没有任何周期性痕迹会出卖它。 冻结后的帧还可以独立于原始素材进行移调， 因此把同一个和弦在多个音高上冻结，就能得到一整层铺底。 由于冻结捕捉的是一个瞬间而非一段时长，选择位置至关重要—— 冻结音符的起音会得到明亮、带棱角的延续音； 冻结尾音则从同一素材得到柔和空洞的声音。

Throat

Vowel formant resonator

元音共振峰谐振器

Built on its own STFT engine rather than wrapping a stretch library with resonators bolted on, which is what lets the vowel character be part of the analysis instead of a filter placed after it. Throat models the vocal tract — the resonant cavity that turns a raw glottal buzz into a recognisable vowel — and lets you move through vowel space on material that never had a voice in it. A pad becomes a choir, a drum loop starts speaking. Because the formants are imposed spectrally rather than by band-pass filtering, they survive on sources with sparse harmonic content where a filter bank would simply find nothing to resonate. Distinct from Spectral Freeze, which is a sustain tool: this is a character tool, and the two chain well together.

构建于自研 STFT 引擎之上，而非在拉伸库外面套一层共振器， 这使得元音特征成为分析过程的一部分，而不是事后附加的滤波器。 Throat 对声道建模——正是这个共鸣腔体 把原始声门振动变成可辨识的元音—— 并让你在原本毫无人声成分的素材上于元音空间中游走。 铺底变成合唱，鼓循环开始"说话"。 由于共振峰是在频谱域施加而非依靠带通滤波， 因此在谐波成分稀疏的素材上依然成立， 而滤波器组在这类素材上根本找不到可共振的内容。 它与 Spectral Freeze 不同：后者是延音工具，这是音色工具，两者串联效果很好。

Vector Field

3-module vector modulation

三模块矢量调制

Four filter and distortion modules, each driven by three LFOs assigned to X, Y and Z axes, with independent delays per module. The vector-field metaphor is literal: each module is a point moving through a three-dimensional modulation space, and its position determines cutoff, drive and delay simultaneously. Because all three axes come from independent LFOs at different rates, the modules trace paths that never repeat exactly — related motion rather than four separate LFOs going their own way, which is what keeps it coherent instead of chaotic. Set the rates in simple ratios and you get rhythmic interlocking; set them irrationally and it evolves for minutes without obviously cycling. Per-module delay means the four voices smear across time as well as spectrum.

四个滤波 / 失真模块，每个由分配到 X、Y、Z 三个轴的三个 LFO 驱动， 每模块还有独立延时。这个"矢量场"比喻是字面意义上的： 每个模块都是在三维调制空间中运动的一个点， 其位置同时决定截止频率、驱动量与延时。 由于三个轴来自速率各异的独立 LFO，各模块描绘的轨迹永不精确重复—— 它们是彼此关联的运动，而非四个各行其是的 LFO， 这正是它保持连贯而不至于混乱的原因。 把速率设成简单整数比会得到节奏性的互锁； 设成无理数比则可以演化数分钟而不明显循环。 逐模块延时意味着这四个声部不仅在频谱上、也在时间上彼此错开。

Soft Clipper

Soft / hard clipper

软 / 硬削波器

A saturation stage that rounds peaks instead of shearing them. A hard clipper replaces everything above the ceiling with a flat line — hundreds of consecutive samples at exactly full scale — which generates high-order odd harmonics and reads as harsh digital breakup. This rounds the top of the waveform through a smooth curve, so the transition into limiting is gradual and the harmonics it adds fall off quickly with order: warmth rather than damage. Output is mathematically guaranteed to stay within the ceiling, so it doubles as a safety stage. Placed before a limiter it removes the sharpest peaks so the limiter works less and pumps less; used alone with drive it thickens a source without noticeably reducing dynamics.

一个把峰值削圆而非切平的饱和级。硬削波器会把天花板以上的一切替换为一条水平线—— 数百个连续采样恰好停在满刻度——这会产生高阶奇次谐波，听感上就是刺耳的数字破音。 而这里通过一条平滑曲线把波形顶部削圆，因此进入限制的过程是渐进的， 所产生的谐波随阶次迅速衰减：得到的是温暖而非损伤。 输出在数学上保证不超过天花板，因此它同时可以作为安全级使用。 置于限制器之前，它会先削掉最尖锐的峰值，使限制器工作更轻、"喘息"更少； 单独配合驱动使用时，则能在几乎不损失动态的前提下让音色变厚。

Reliquary

Banded-waveguide resonator

带状波导谐振器

A spectral banded waveguide. Where a classic recycling delay loops one line and gets a single compromised pitch out of it, Reliquary runs a bank of short feedback loops — one per harmonic partial of a tracked fundamental. Each loop is fed a band-pass slice of your own audio, so it recycles the part of your signal sitting near that partial rather than synthesised noise or a string model. Each has a fractional delay length tuned by a first-order allpass interpolator, so every partial is dead in tune at any pitch — the thing integer and linearly-interpolated loops both get wrong. Each partial carries its own decay, so lows can ring while highs damp like a real string, or the reverse for something metallic. One CHARACTER macro bends the partial ratios from perfectly harmonic through stretched and inharmonic to dense clusters: string to bell to drone on a single knob, with tuning, spectrum and decay stated independently underneath.

频谱带状波导。经典的循环延时只回收一条延时线，只能得到一个充满妥协的音高； 而 Reliquary 运行一整组短反馈环路——每个对应所跟踪基频的一个谐波分音。 每个环路被送入的是你自己音频的带通切片， 因此它回收的是你信号中位于该分音附近的成分，而不是合成噪声或弦模型。 每个环路都有由一阶全通插值器调谐的小数延时长度， 因此任何音高下每个分音都精确在调上—— 这恰是整数延时环路与线性插值环路都做不好的地方。 每个分音有独立衰减，因此低频可以延续而高频衰减，像真实的弦； 反过来则得到金属质感。单一 CHARACTER 宏控件可把分音比例从完全谐波， 经拉伸、失谐，一直推向致密簇：一个旋钮走完从弦、到钟、到嗡鸣， 而调音、频谱与衰减在其下各自独立设定。

GrainMate FX

Real-time granular effect

实时颗粒效果器

The GrainMate engine pointed at your live input. It captures incoming audio into a rolling four-second buffer and granulates it on the fly — pitch-shift, smear, stretch and scatter, fully wet / dry mixable. Because the buffer is continuously refreshed, the grains always come from what you are playing now, so it tracks a performance instead of replaying a sample. Freeze snapshots the buffer and grains it indefinitely, turning any passing moment into a sustained texture. Feedback routes output back into the capture, so grains re-granulate themselves and the texture evolves rather than repeating — the difference between a granular effect and a granular instrument. The scan position locks to tempo, so the read head moves in time with the track. A live view shows the capture buffer, the read head and each grain as it fires.

把 GrainMate 引擎对准你的实时输入。它将传入音频写入一个滚动的四秒缓冲区并即时颗粒化 ——移调、涂抹、拉伸与打散，干 / 湿可自由混合。 由于缓冲区在持续刷新，颗粒始终来自你当下正在演奏的内容， 因此它跟随的是演奏本身，而不是回放一个采样。 Freeze 冻结当前缓冲并无限颗粒化， 把任何一个转瞬即逝的瞬间变成持续的纹理。 Feedback 把输出送回捕获端， 使颗粒对自身再次颗粒化，纹理不断演化而非重复—— 这正是"颗粒效果器"与"颗粒乐器"之间的区别。 扫描位置锁定到速度，因此读头随曲目节奏移动。 实时视图显示捕获缓冲、读头，以及每个颗粒触发的瞬间。

QuadraTape

4-band multiband saturator

四段多频段饱和器

A multiband saturator: the signal is split into four bands by a Linkwitz-Riley crossover tree — three crossovers, default 191 Hz / 1.4 kHz / 3.1 kHz — with allpass reconstruction, so with every band clean the four sum back to a magnitude-flat copy of the input. The default patch is genuinely transparent; you are only setting band borders until you push something. Then each band saturates through one of five shapers: TAPE (soft symmetric tanh compression), TUBE (asymmetric, even-harmonic warmth), DIST (cubic clipper, aggressive odd harmonics), AMP (two-stage cascaded squash) and DEC (decimator — sample-hold plus quantise). Per band: drive, a downward gate, output trim and its own dry / wet mix, with a global mix and output on top. Distort the low end without touching the cymbals, or crush only the top band.

多频段饱和器：信号由 Linkwitz-Riley 交叉树分成四个频段—— 三个分频点，默认 191 Hz / 1.4 kHz / 3.1 kHz——并配合全通重建， 因此当所有频段都保持干净时，四段求和后是输入信号的幅度平坦副本。 默认预设是真正透明的；在你推动任何一段之前，你只是在设定频段边界。 随后每个频段可通过五种整形器之一进行饱和：TAPE（柔和对称 tanh 磁带压缩）、 TUBE（非对称、偶次谐波的电子管温暖感）、DIST （三次方削波器，强烈的奇次谐波）、AMP（两级级联箱体挤压）与 DEC（降采样器——采样保持加量化）。每段独立设置：驱动、向下噪声门、 输出微调及各自的干 / 湿混合，之上还有全局混合与输出。 可以只让低频失真而不碰镲片，或只压碎最高频段。

Force Field Delay

Particle-field delay — new in v1.9.5

力场延时 —— v1.9.5 新增

Most delays give you a time and a feedback amount. This one treats every echo as a particle moving through a field, and gives you the field. Each tap has an age, and the force acts on it continuously as it ages — bending its delay time, its stereo position, its feedback path. The repeats do not just decay; they travel. Thirteen field types, each a different force law: Vortex runs time modulation in quadrature between channels so echoes orbit rather than oscillate; Magnetic applies force perpendicular to motion with heavy cross-feed; Harmonic is a damped spring whose swing settles with tap age; Charge accelerates taps so repeats crowd together; Lennard-Jones uses the intermolecular potential — repulsive close in, attractive further out — so taps push apart then pull back together. Plus Wind, Drag, Turbulence, Texture, Curve Guide, Boid, Fluid Flow and a flat Force reference. FALLOFF and POWER shape how force decays with distance; MIN / MAX DIST bound where it acts at all. Tempo-synced across the same 37-step ladder as the Delay, or free in milliseconds.

大多数延时只给你时间和反馈量。这一个把每次回声当作在场中运动的粒子， 并把这个"场"交给你。每个抽头都有年龄，力会随着它变老而持续作用于它—— 弯曲它的延时时间、立体声位置与反馈路径。回声不只是衰减，它们在移动。 十三种场类型，每种对应不同的力学定律：Vortex（涡旋） 让两声道的时间调制呈正交关系，因此回声是在环绕而非振荡； Magnetic（磁力）施加垂直于运动方向的力并伴随强交叉反馈； Harmonic（谐振）是一根阻尼弹簧，摆幅随抽头年龄而平息； Charge（电荷）使抽头加速，让回声越挤越密； Lennard-Jones 采用分子间势能——近距排斥、远距吸引—— 因此抽头先推开再彼此拉回。此外还有 Wind、Drag、Turbulence、Texture、 Curve Guide、Boid、Fluid Flow，以及作为参照的平直 Force。 FALLOFF 与 POWER 塑造力随距离的衰减方式；MIN / MAX DIST 限定力的作用范围。 可同步到与 Delay 相同的 37 级阶梯，或以毫秒自由设定。

Modal

Parallel modal resonator bank — new in v1.9.5

并联模态谐振器组 —— v1.9.5 新增

Modal synthesis represents a resonant object as a sum of second-order modes. Where Reliquary is a banded waveguide — tuned delay loops, limited to twelve partials by memory — this is a bank of resonators, costing a handful of registers each and scaling to 512 modes. Three decisions make it sound like an object rather than a filter bank: each mode is a coupled-form rotation, so frequency IS the rotation angle and decay IS the radius — no biquad coefficients to lose precision in when poles sit microscopically close to the unit circle at long decays; every mode carries a zero as well as its poles, which is what fills the response between resonances instead of leaving comb notches; and injection is energy-normalised with a bounded loss, so the ring is usable on transients without a sustained tone running away. Stiff-string inharmonicity (f_k = k·f₀·√(1+B·k²)) sweeps from harmonic through bell to dense metallic clusters. STRIKE is a transient shaper on the excitation — measured, it sharpens a kick by 35 dB while ducking sustained material, which is what turns a drone into something that reads as struck. Optional MIDI tuning, per-mode stereo spread, and a live display of the actual mode layout on a log axis.

模态合成将一个共鸣物体表示为若干二阶模态之和。Reliquary 是带状波导—— 调谐延迟环路，受内存限制只能有十二个分音；而这是一组谐振器， 每个仅需少量寄存器，可扩展至 512 个模态。 三项决策让它听起来像一个物体而非滤波器组：每个模态采用耦合形式旋转， 频率即旋转角、衰减即半径——长衰减时极点极度贴近单位圆，而这里没有会损失精度的 biquad 系数；每个模态除极点外还带有一个零点，这正是填补谐振之间响应的关键， 否则会留下梳状凹陷；输入注入采用能量归一化并配合有界损耗， 因此瞬态激励下的振铃可用，而持续音也不会失控。 刚性弦失谐（f_k = k·f₀·√(1+B·k²)）可从谐波扫过钟声直至致密的金属簇。 STRIKE 是作用于激励信号的瞬态整形器——实测可将底鼓锐化 35 dB， 同时压低持续素材，这正是把嗡鸣变成"被敲击感"的关键。 可选 MIDI 跟随音高、逐模态立体声展开，并以对数坐标实时显示真实的模态分布。

Linear Phase EQ

Brick-wall high-pass / low-pass

砖墙式高通 / 低通

Surgical frequency shaping without phase dispersion. A minimum-phase filter — every analogue EQ and most digital ones — shifts different frequencies by different amounts of time, so cutting the low end of a drum bus moves the low frequencies of the attack transient relative to the highs, smearing it. The linear-phase kernel introduces zero group-delay variation: every frequency is delayed identically, so the waveform's shape survives the filtering intact. That matters most where phase relationships carry the information — parallel processing, multi-mic sources, mastering, anywhere the filtered signal will be summed against something unfiltered. High-pass and low-pass in one unit with adjustable corner frequencies, fully automatable. The trade is latency, which the DAW compensates for automatically per slot.

精确的频率塑形，不产生相位色散。最小相位滤波器——所有模拟 EQ 与大多数数字 EQ—— 会让不同频率产生不同的时间偏移，因此对鼓总线做低切时， 起音瞬态的低频会相对高频发生位移，从而被涂抹。 线性相位内核引入零群延迟变化： 所有频率被延迟相同的时间，因此波形形状在滤波后完好无损。 这在相位关系承载信息的场合最为关键——并行处理、多话筒录音、母带， 以及任何滤波后信号需要与未滤波信号求和的场景。 高通与低通合为一体，拐角频率可调，完全支持自动化。 代价是延迟，而 DAW 会为每个插件位自动补偿。

REQ6

6-band parametric EQ

6 段参量均衡

Six fully parametric bands, each independently switchable between ten filter types: bell, low shelf, high shelf, high-pass, low-pass, notch, and resonant low / high shelves that add a bump at the corner frequency. Bands 1 and 6 additionally offer steep high-pass and low-pass — a four-stage Butterworth cascade giving roughly 48 dB/octave, double a standard filter, with a maximally-flat passband and no peak at the corner. That is the difference between rolling off the subs and genuinely removing them. Every band has its own frequency, gain and Q, all fully automatable, drawn as a live frequency-response curve so you can see the sum of what the six bands are doing rather than reading six sets of numbers. Coefficients are snapshotted under lock at the top of each block, so automating a sweep stays glitch-free.

六个完全参量的频段，每段可独立切换为 十种滤波器类型： 钟形、低架、高架、高通、低通、陷波，以及在拐角频率处带有隆起的谐振低 / 高架。 第 1 和第 6 频段还额外提供陡峭高通与低通—— 四级 Butterworth 级联，约 48 dB/倍频程，是标准滤波器的两倍， 通带最大平坦且拐角处无峰。这正是"衰减低频"与"真正去除低频"之间的区别。 每个频段都有独立的频率、增益与 Q，全部支持自动化，并以实时频响曲线绘制， 因此你看到的是六段叠加后的总体效果，而不是六组数字。 滤波系数在每个音频块开头加锁快照，因此自动化扫频始终无爆音。

Orbit

Auto-panner with drawable motion curves

可绘制运动曲线的自动声像器

New in v1.9.6. A tempo-synced auto-panner where the motion is drawn rather than chosen — a 64-point curve per band replaces the fixed sine of a conventional panner, with three bands so the bass can hold centre while the top moves. Three things separate it from the field. First, the panning law is a control, not a hardcoded assumption: an auto-pan crosses the centre repeatedly, and the law decides what happens there — linear is constant-amplitude (correct for a mono fold-down but dips ~3 dB through centre on speakers), −3 dB is constant-power (even on speakers, bumps in mono), −4.5 dB is the broadcast compromise that is near-flat in both. On a fast pattern that choice is an audible amplitude pulse on every crossing. Second, Haas panning is metered, not merely offered: delay-based panning is the same signal at two times, which is the definition of a comb filter when summed, so a live correlation meter reads the actual output while you dial it. Third, a bass-lock mono-below control sits independent of the curves, because low-frequency panning is never wanted and is easy to introduce by accident. An envelope follower can drive the motion from the material, and curves live in the plugin state rather than as parameters — 64 points across 3 bands would otherwise be 192 automation lanes for something you draw once; an automatable four-slot curve selector switches between them instead.

v1.9.6 新增。一款速度同步的自动声像器，其运动轨迹是 画出来的，而非选出来的——每个频段 64 个控制点的自定义曲线， 取代了传统声像器固定的正弦波形；三个频段各自独立，因此低频可以牢牢锁在中央， 而高频自由游走。三点让它区别于同类产品。其一，声像定律是一个可调控制项， 而非写死的假设：自动声像会反复穿越中心点，而定律决定了穿越时发生什么—— Linear（线性）为恒幅度（适合单声道折叠，但在音箱上过中心时会下陷约 3 dB）、 −3 dB 为恒功率（在音箱上均匀，但单声道时会隆起）、 −4.5 dB 则是广播领域的折中方案，两种情况下都接近平坦。 在快速律动中，这个选择意味着每次穿越中心都会产生可闻的音量脉动。 其二，哈斯（Haas）声像带有实时计量，而不只是提供功能： 基于延时的声像本质上是同一信号的两个时间副本，求和后即构成梳状滤波器， 因此内置的相关性表会实时读取真实输出，让你在调节时就能看到代价。 其三，Bass Lock（低频锁定）低频单声道化控制独立于所有曲线之外， 因为低频声像几乎从不是你想要的，却极易被无意引入。 包络跟随器可让运动跟随素材本身起伏；曲线保存在插件状态中而非参数中—— 3 个频段 × 64 个点若作为参数将产生 192 条自动化轨道，而这只是你画一次的东西； 取而代之的是一个可自动化的四槽曲线选择器。

Expanse

Multiband stereo enhancer

多频段立体声增强器

New in v1.9.6. The classic hardware stereo expander (the SPL Vitalizer lineage) crossfeeds inverted opposite-channel content — L′ = L − kR, R′ = R − kL — and its celebrated property is provable mono compatibility, since L′ + R′ = (1 − k)(L + R): a mono fold-down is a level change and nothing else. But rewrite that same operation in M/S and the flaw appears — mid × (1 − k) against side × (1 + k). The mid falls as the side rises; at k = 0.8 the centre is 14 dB down. Every review of the hardware reports the same thing: push it and the centre drops out. That is not an artefact, it is the operation working as designed. Expanse does four things the original could not. It compensates the centre, restoring mid to unity after the crossfeed, so width opens without hollowing out the vocal — which alone changes how far the control can usefully be pushed. It runs per band, a Linkwitz-Riley crossover tree splitting into four bands each with its own width, so bass stays mono and solid while the top opens. It decorrelates rather than delays: a cascade of allpass sections on the side signal only spreads phase across frequency at flat magnitude, widening the image with no comb notches and leaving the mono sum untouched. And it measures itself — a running correlation coefficient tells you whether what you just did is mono-safe, rather than leaving you to find out on a club system.

v1.9.6 新增。经典硬件立体声扩展器（SPL Vitalizer 一脉）的做法， 是将反相的对侧声道内容交叉馈送——L′ = L − kR、R′ = R − kL—— 其广受称道的特性是可被证明的单声道兼容性，因为 L′ + R′ = (1 − k)(L + R)：单声道折叠只是电平变化，别无其他。 但把同一运算改写为 M/S 形式，缺陷立刻显现——Mid × (1 − k) 对 Side × (1 + k)。中间信号随着侧信号升高而衰减；当 k = 0.8 时， 中心已经下降了 14 dB。所有关于该硬件的评测都指出同一件事： 推得太狠，中心就会塌陷。这并非瑕疵，而是该运算按设计工作的必然结果。 Expanse 做到了原始电路无法做到的四件事。它补偿中心， 在交叉馈送之后把 Mid 恢复至单位增益，因此声场展宽不会掏空人声—— 仅此一点就改变了这个控制项可被有效推进的幅度。它逐频段运行， 由 Linkwitz-Riley 分频树切分为四个频段，各自拥有独立的宽度控制， 因此低频保持单声道且扎实，而高频得以打开。它采用去相关而非延时： 仅对侧信号施加一组级联全通滤波器，在幅频响应平坦的前提下将相位随频率铺开， 从而在没有任何梳状陷波的情况下展宽声像，单声道求和完全不受影响。 并且它自我度量——实时相关系数会告诉你刚才所做的处理是否单声道安全， 而不是让你到俱乐部音响上才发现问题。

The Top Rail Glitch & Shiftermate, always one keystroke away

顶部工具栏 Glitch 与 Shiftermate —— 一键即达

Plus a global Pitch slider in the same toolbar — a project-wide transposition in semitones that re-pitches every audio region in real time without altering its tempo, when Stretch mode is enabled.

此外,同一工具栏中还提供全局 Pitch 滑杆——可在整个工程范围内以半音为单位进行实时移调(当 Stretch 模式启用时),在不改变原始时长的情况下对所有音频区域进行重音高处理。

Per-Channel Superpowers five switches per track, PRE or POST the chain

每声道超能力 每轨五个开关,插件链前或链后

Every audio and group track has a compact row of seven switches in its channel strip — five effect toggles bracketed by chain-position selectors:

每条音频轨和编组轨在其通道条上都有一排紧凑的七个开关—— 五个效果开关被链路位置选择器包夹:

HarmonicBufferPitch (P button)

HarmonicBufferPitch(P 按钮)

Click P to engage a per-track harmonic buffer pitch: a 1/16-note-locked chunk repeater whose chunk size sets the perceived fundamental. Capture the first N samples of every 1/16th, repeat that captured chunk for the rest of the 1/16th with click-free fades, leave a small empty tail rather than overshoot the bar grid. Range 16 Hz to 16744 Hz — ten chromatic octaves. A separate ♪ button toggles the value display between continuous Hz and chromatic notes (440 Hz ↔ A4); scroll-wheel always moves by one semitone, Shift+scroll by an octave. The Hz parameter is fully automatable — draw a curve to sweep the pitch across ten octaves over a phrase, or step it up the scale every bar.

点击 P 启用逐轨 HarmonicBufferPitch:一个锁定到 1/16 音符的块重复器, 其块长度决定听感上的基频。在每个 1/16 内捕获前 N 个采样, 用无咔嗒的淡入淡出在该 1/16 余下时间内重复这块,并留出一小段空尾以防越过小节网格。 范围 16 Hz 到 16744 Hz——十个半音八度。一个独立的 ♪ 按钮在连续 Hz 与半音音名(440 Hz ↔ A4)之间切换显示; 滚轮始终以一个半音步进,Shift+滚轮则按八度。Hz 参数完全可自动化 ——画一条曲线让音高在一个乐句内扫过十个八度,或者每小节按音阶上行一级。

Per-track Glitch and Shifter (G, S buttons)

逐轨 Glitch 与 Shifter(G、S 按钮)

The same Glitch and Shiftermate algorithms from the master toolbar, but as independent per-track instances. The master toolbar dialogs (Ctrl+G, Ctrl+M) tune the sound; the per-track G / S buttons decide which channels run the effect. Both toggles are individually automatable as step functions — draw square waves on bar boundaries to chop the effect in and out per phrase.

与主工具栏完全相同的 Glitch 与 Shiftermate 算法,但作为每轨独立的实例运行。 主工具栏对话框(Ctrl+G、Ctrl+M)用来调音色;逐轨的 G / S 按钮决定哪些声道运行效果。 两个开关都可作为阶梯函数独立自动化——在小节边界上画方波,即可按乐句切入切出。

Quick LC / HC filters

快速 LC / HC 滤波器

One-click low-cut and high-cut (Butterworth biquad, Q=1/√2) on every track with adjustable Hz. Click a Hz value to drag-edit, mouse-wheel to scroll by 6 % per tick (≈ semitone resolution), or double-click for direct numeric entry.

每条轨道上的一键低切与高切(Butterworth 双二阶,Q=1/√2),Hz 可调。 点击 Hz 数值进行拖动编辑,滚轮每格变化 6 %(约一个半音的分辨率), 双击则直接键入数值。

PRE / POST routing

PRE / POST 路由

The whole effect group sits either before or after the track's plugin chain — toggle with the PRE / POST switches at either end of the row. POST is default (effects polish the chain output); PRE feeds glitch / shifter / pitch INTO the chain so your reverbs and EQs hear the warped signal. A coloured frame around the active half keeps the routing answer visible at a glance.

整个效果组要么放在轨道插件链 之前,要么放在 之后 ——通过这一行两端的 PRE / POST 开关切换。默认是 POST(效果对链路输出进行打磨); PRE 则把 glitch / shifter / pitch 喂进链路,让你的混响与 EQ 听到这些扭曲后的信号。 激活的一半周围有彩色边框,确保路由方向一眼可辨。

Kick & Bass Behavior invisible sidechain, zero routing

Kick 与 Bass Behavior 隐形侧链,无需任何路由

The classic psytrance pump — kick ducks everything, bass ducks the rest — built straight into the mixer, with no sidechain plugins, sends or routing. Right-click a channel → Add Behavior and tag it as Kick or Bass; the engine does the rest from that channel's own onsets.

经典的 psytrance 律动泵感——kick 压低一切,bass 压低其余——直接内建在混音器里, 无需任何侧链插件、发送或路由。右键某条音轨 → Add Behavior, 将其标记为 Kick 或 Bass;引擎会根据该音轨自身的起音点完成余下的一切。

It's completely non-destructive: the duck is applied only on the master render (live playback and master export), never written into your regions, and it steps aside when a single channel is soloed for an isolated bounce. Depth, attack, recovery, and a pre-onset pre-cut hold are all tunable in Behavior Engine (Kick / Bass ducking) — and every parameter is automatable.

它完全非破坏性:压低只作用于主输出渲染(实时播放与主输出导出), 绝不写入你的区域;当你单独 solo 某条音轨做隔离导出时,它会自动让路。 深度、起音、恢复,以及起音前的 pre-cut 静音保持,都可在 Behavior Engine(Kick / Bass ducking) 中调节——且每个参数都可自动化。

Drag & Drop Anything including images

拖入任何东西 包括图像

The timeline accepts the usual audio formats (WAV, AIFF, FLAC, OGG, MP3) by drag-and-drop — but it also accepts image files. Drop a JPG, PNG, or BMP onto a track and HarmonicDAW sonifies it: pixels are scanned row-by-row, brightness becomes amplitude, linear-interpolation expands each pixel into multiple samples, and the result is written as a 24-bit stereo WAV at the project sample rate. A new audio track is auto-created (named after the image), the resulting region drops in at the cursor position with default fades applied, ready to be edited like any other audio. Same shortcut works for dropping images onto an existing audio track — the image is converted in place into that track's chain.

时间线通过拖放支持常见音频格式(WAV、AIFF、FLAC、OGG、MP3),但它同时也接受 图像文件。将一个 JPG、PNG 或 BMP 拖到轨道上,HarmonicDAW 就会把它声音化: 像素逐行扫描,亮度变成幅度,线性插值把每个像素扩展为多个采样,结果以工程采样率写为 24 位立体声 WAV。一条新的音频轨会自动创建(以图像命名),生成的区域落在光标位置并自动应用 默认淡入淡出,可像任何其他音频一样编辑。同一个快捷方式也适用于把图像拖到已有的音频轨上 ——图像将就地转换并进入该轨道的链路。

Useful for turning glitch art, photos of waveforms, fractals, or text scans directly into sound material. Combined with the per-channel Glitch / Shifter / harmonic-pitch effect group, an arbitrary JPG becomes the seed of a full track in seconds.

可用于把故障艺术、波形照片、分形或文本扫描直接转化为声音素材。 再配合逐声道的 Glitch / Shifter / harmonic-pitch 效果组,任意一张 JPG 几秒钟就能成为一整条轨道的种子。

Architecture specs & internals

架构 规格与内部细节

Acquisition

获取

Try the DEMO for 14 days, free of charge — full feature set, time-limited by the embedded TrialManager. When you're ready to lift the limit, buy PRO for $35 in PSYKO on BNB Smart Chain: connect a BEP-20 wallet (MetaMask, Trust, etc.), confirm the transaction, your download starts automatically. No accounts, no email confirmations, no subscriptions, no refund department.

免费试用 DEMO 14 天——完整功能集,由内置的 TrialManager 进行时间限制。 准备解除限制时,在 BNB 智能链上以 $35 等值的 PSYKO 购买 PRO: 连接一个 BEP-20 钱包(MetaMask、Trust 等),确认交易,下载自动开始。 没有账号注册,不发邮件确认,不订阅,不设退款部门。