Erwin Smith Voice Impression Guide
Commander Erwin Smith delivers the most kinetically charged speech in Attack on Titan with a voice that feels like a natural force — controlled, resonant, and capable of rallying thousands to certain death. Whether you want to recreate that “WE GIVE OUR HEARTS!” intensity for Discord roleplay, cosplay events, streaming, or AI voice content, this guide breaks down the complete acoustic anatomy of Erwin’s voice, maps out specific DSP settings, covers physical training drills, and walks through an AI voice cloning workflow on Windows.
TL;DR
- Erwin’s voice is a controlled low-baritone with exceptional chest resonance, deliberate pacing, and explosive dynamic range on key phrases — not a deep character-voice gimmick but a disciplined performance craft.
- The Japanese dub (Daisuke Ono) sits around 100–120 Hz fundamental with crisp consonant articulation; the English dub (J. Michael Tatum) is warmer and slightly fuller at 105–125 Hz.
- DSP settings: −2 to −4 semitones pitch shift, mild chest formant emphasis, moderate projection compression with fast attack and slow release.
- Physical drills — ribcage breathing, vowel elongation, sustained projection — bridge the gap DSP cannot cover.
- AI voice cloning handles the fine-grained vocal character that pitch shift alone cannot reproduce, with sub-300ms latency on a mid-range GPU.
- VoxBooster on Windows supports AI model import, WASAPI routing, and Discord/OBS integration with no kernel driver required.
Who Is Commander Erwin Smith?
Commander Erwin Smith is the 13th Commander of the Survey Corps in Attack on Titan, the manga series by Hajime Isayama and the Wit Studio / MAPPA anime adaptation. He is defined by a paradox: unflinching strategic ruthlessness paired with genuine compassion for the soldiers who follow him. His speeches — particularly the charge on the Beast Titan in Season 3 — are among the most emotionally overwhelming moments in the series precisely because his voice makes you believe in the mission even when the math is clearly fatal.
That believability is not accidental. Both the Japanese voice actor Daisuke Ono and the English dub voice actor J. Michael Tatum built Erwin’s voice around specific performance choices that translate into identifiable acoustic properties you can analyze, practice, and replicate.
The Acoustic Anatomy of Erwin’s Voice
Before touching any software settings, understanding what you are trying to recreate prevents you from chasing the wrong parameters.
Fundamental Range and Chest Placement
Erwin’s baseline speaking voice sits in the low-baritone range — approximately 100–120 Hz in the Japanese dub performance by Daisuke Ono, and 105–125 Hz in J. Michael Tatum’s English version. This is not an extreme bass voice. The power does not come from subterranean frequency; it comes from chest resonance and placement.
The key distinction: Erwin projects from a relaxed, low chest position rather than a tightened throat. This produces a rounded, full fundamental with clean overtones rather than the raspy, constricted quality that a forced “deep voice” attempt creates. If your attempt sounds tense or strained, you are working from the throat rather than the chest.
Deliberate Articulation and Pacing
Erwin speaks with conscious control over every word in dialogue scenes. His articulation is crisp — consonants are clean and fully pronounced, not swallowed. His pacing is deliberate: slightly slower than natural speech in strategic moments, with clear rhythmic emphasis on key nouns and commands.
This articulation pattern is one of the hardest aspects to capture because it requires conscious performance discipline, not just audio processing. Software can shift your pitch; it cannot insert the millisecond pause before “humanity” or the drop in volume that Ono uses to devastating effect before Erwin’s rallying cry climax.
The Rally Cry Dynamic Range
The sequence that defines the voice — the charge speech in Season 3, Episode 17 — demonstrates extraordinary dynamic range. Erwin starts at a controlled forte, builds methodically through a crescendo that compresses the rhythm of the sentences, then releases into a full-voiced forte on “WE GIVE OUR HEARTS!” where the voice opens and expands rather than straining upward.
This is the opposite of screaming. The volume increases while tension decreases — chest opens, projection expands, the voice gets fuller rather than thinner. Any compression or limiting in your processing chain needs fast-attack / slow-release characteristics to preserve this dynamic expansion rather than flattening it.
DSP Settings for the Erwin Voice Effect
DSP-only processing gets you into Erwin’s territory quickly with no model training required. These settings work in any Windows real-time voice changer that supports pitch shift, EQ, and compression.
Pitch Shift
| Starting Voice Type | Target Semitones |
|---|---|
| Tenor (typical male) | −3 to −4 semitones |
| Baritone (typical male) | −1 to −2 semitones |
| Bass (natural) | 0 to −1 semitone |
| Female soprano | −9 to −11 semitones |
| Female mezzo | −7 to −9 semitones |
Use a high-quality pitch shift algorithm — formant-preserving modes produce a much more natural result than basic pitch transposition, which creates a chipmunk-reversal artifact at large shifts.
Formant Targeting
Enable chest formant emphasis or a “male voice” formant preset if your software offers it. The target is a slight lowering of the first formant (F1) and a modest lowering of the second formant (F2), which thickens the vowel resonance and adds the characteristic “chest weight” to the voice.
If you have a parametric EQ available, apply a gentle +2 to +3 dB boost around 150–250 Hz (chest body), a slight −1 dB cut around 3–4 kHz (reduces harshness), and a gentle high-frequency roll-off above 10 kHz. This keeps the voice warm and authoritative rather than harsh or bright.
Compression
Erwin’s voice has narrow dynamic range in calm speech — authority implies control. Use a compressor with:
- Ratio: 3:1 to 4:1
- Attack: 5–10 ms (fast enough to catch peaks without killing transients)
- Release: 100–200 ms (slow enough to preserve the expansion dynamic on phrases)
- Threshold: set so gain reduction activates on peaks, leaving normal speech largely unprocessed
- Makeup gain: +1 to +2 dB after compression to restore presence
Avoid over-compressing. Erwin’s voice uses its dynamic range for effect. A heavily compressed voice loses the strategic variability that makes the character feel calculated rather than robotic.
Optional: Presence Boost
A gentle boost at 1–2 kHz adds “projection” — the quality of a voice that carries across a large space. Military commanders and trained orators all develop this through resonance placement; a soft +1.5 dB shelf at 1 kHz approximates it electronically.
Physical Training Drills
DSP closes the gap but cannot replace the vocal quality that comes from proper technique. These drills directly develop the chest resonance, breath control, and articulation that define Erwin’s performance style.
Ribcage Breathing
Erwin’s volume comes from breath support, not throat tension. Lie on your back, place one hand on your chest and one on your abdomen. Breathe in slowly, pushing both hands up. This activates the diaphragm-supported breathing pattern. Practice speaking sustained vowels (“AH,” “OH”) while maintaining this low-body sensation. The goal is to feel vibration in your sternum rather than your throat.
Practice duration: 10 minutes daily for two weeks to establish the muscle memory pattern.
Vowel Elongation Drill
Take any of Erwin’s iconic lines — “If you trust in me, follow!” — and practice it at half speed, holding each stressed vowel for twice its natural duration. This forces your articulators into full, open positions rather than the lazy vowel reduction that characterizes casual speech. After the slow version feels comfortable, return to normal speed. The openness usually carries over.
Sustained Projection
Stand facing a wall at five meters distance. Speak Erwin lines at conversational volume — not loud — with the intent of making the sound reach the wall clearly. This develops the resonance placement that makes a voice carry without shouting. Gradually increase to ten meters. The exercise builds the chest-forward projection quality without the strain of yelling.
The Phrase Architecture Drill
Erwin builds pressure through repetition and rhythmic stacking. Identify the structural pattern in his rally speech: statement → intensification → release. Practice delivering any three-sentence sequence using this architecture, with deliberate slower pacing on the final beat before the release. This builds the performance instinct that software cannot insert.
AI Voice Cloning Workflow
For the highest fidelity Erwin voice impression, AI voice cloning captures the specific timbre, resonance pattern, and micro-articulations that pitch shift cannot reproduce.
Source Audio Preparation
Collect 15–30 minutes of clean Erwin dialogue. The critical requirement is isolation — the AOT soundtrack layers music and sound effects heavily over most scenes, and training on contaminated audio degrades the model quality significantly.
For the Japanese voice (Daisuke Ono), isolated drama CD recordings or clean audio rips from Blu-ray editions offer the cleanest source. For the English voice (J. Michael Tatum), isolated dub recordings without the Japanese audio track give the best separation. Community audio repositories often have pre-isolated versions.
Segment the audio into clips that cover Erwin’s emotional range: calm strategic dialogue, moderate command authority, and peak rally intensity. A model trained only on conversational tone will struggle to reproduce the rally cry dynamic without distortion.
Preprocessing
Before training:
- Trim silence at clip boundaries (leave 0.2–0.5 s natural breath pauses)
- Normalize to −18 LUFS integrated loudness
- High-pass filter at 80 Hz to remove room rumble
- Check for any remaining music bleed using spectral analysis and discard contaminated clips
Model Training and Import
Train the model through an AI voice conversion tool that supports custom model import. Standard training runs at 50,000–200,000 steps depending on data volume; 15–20 minutes of clean audio typically reaches usable quality at 50,000–80,000 steps and peak quality near 150,000 steps.
Once trained, export the model in the tool’s native format. VoxBooster on Windows supports direct AI voice model import — drop the model file into the Models folder in the VoxBooster data directory, restart the application, and it appears in the voice selection dropdown. No Python environment, no manual configuration, no kernel driver. The sub-300ms inference latency on a GTX 1060-class GPU is fast enough for live Discord conversations.
Combining DSP and AI Conversion
For best results, apply the DSP pitch shift and EQ settings described above as pre-processing before the AI voice conversion layer. This pre-conditions your input voice closer to Erwin’s range, reducing the conversion distance the model has to bridge and improving output naturalness. An 8–10 dB noise gate before the conversion stage also reduces background noise bleed that AI models can artifact-ify into unusual timbres.
Setting Up for Discord and OBS
Discord Configuration
- Install VoxBooster and configure your Erwin settings (DSP chain, or AI model loaded and selected).
- Open Discord → Settings → Voice & Video.
- Under Input Device, select “VoxBooster Virtual Microphone.”
- Disable Discord’s built-in noise suppression and echo cancellation — these algorithms conflict with real-time voice conversion and introduce phase artifacts that degrade the output.
- Set input sensitivity to manual rather than automatic, with the threshold set below Erwin’s projected speaking level.
- Test in a private server or the Discord Echo Test Bot before using in a call.
OBS Configuration
- In OBS, add an Audio Input Capture source.
- Select “VoxBooster Virtual Microphone” as the device.
- In the audio mixer, apply a noise gate filter (close threshold: −50 dB, open threshold: −40 dB) to prevent bleed during silence.
- Apply a small reverb or room simulation filter if you want the “echoing command” quality of Erwin’s outdoor rally scenes — a short pre-delay (15–20 ms) and small room size works without muddying the voice.
- Monitor through headphones during a stream test to confirm the output matches your intent before going live.
Comparison: Japanese Dub vs. English Dub Performance Style
| Characteristic | Daisuke Ono (JP) | J. Michael Tatum (EN) |
|---|---|---|
| Fundamental range | ~100–120 Hz | ~105–125 Hz |
| Vowel quality | More closed, precise | Fuller, rounder |
| Consonant sharpness | Crisper, more military | Slightly softer |
| Emotional coloring | Colder authority | Warmer gravitas |
| Rally cry peak | Explosive forward thrust | Expansive and soaring |
| Pacing | Slightly faster | Slightly more deliberate |
| DSP pitch offset | −3 to −4 semitones (most males) | −2 to −3 semitones (most males) |
Neither is superior — they are different performance interpretations of the same character. The English dub version is often more accessible for Western Discord and streaming audiences; the Japanese version has a sharper military edge that cosplay and competitive communities may prefer.
Using the Erwin Voice for Streaming and Roleplay
Beyond technical recreation, Erwin’s voice works in several community contexts:
Survey Corps Roleplay Servers: The structured command authority of Erwin’s delivery fits perfectly into AOT-themed Discord servers. The voice establishes character presence immediately without needing visual context.
Streaming Reaction Content: The “WE GIVE OUR HEARTS!” phrase is one of the most reaction-content-friendly moments in anime history. A processed recreation of the line on top of the original scene creates genuine entertainment value for viewers familiar with AOT.
Tabletop RPG Sessions: Erwin’s style maps cleanly to military commanders, noble strategists, or any NPC requiring authoritative gravitas. The measured pacing and deliberate articulation read as “important character” across any setting.
Cosplay Events and Conventions: A live voice impression is one of the most memorable elements of any character cosplay. With the DSP-only settings dialed in through VoxBooster, you can run the impression on a Windows laptop without carrying dedicated audio hardware.
Ethics and Content Guidelines
Voice impressions of fictional anime characters for non-commercial fan use occupy a well-established tradition in fan communities. For live interactive use — Discord conversations, gaming sessions, convention appearances — the ethical standard is clear identification when context requires it (no sustained identity deception).
For recorded content, avoid creating content that could be mistaken for official material or that depicts the character making statements inconsistent with the source work in contexts that could mislead casual viewers.
For any commercial use of voice content that closely replicates the actual performance of Daisuke Ono or J. Michael Tatum, consult the relevant character licensing and voice actor rights frameworks before publishing. The creative fan space is wide; the commercial edge requires more care.
Frequently Asked Questions
What makes Erwin Smith’s voice acoustically distinctive from other AOT characters?
Erwin’s voice sits in a controlled low-baritone range with exceptional projection and minimal vocal fry. Unlike Levi’s raspy tension or Eren’s raw intensity, Erwin projects deliberate authority — every word lands with strategic weight, and the resonance comes from chest placement rather than throat tension.
How many semitones do I need to shift my pitch to sound like Erwin?
Most male voices need −2 to −4 semitones to reach Erwin’s fundamental range. Daisuke Ono’s Japanese performance sits around 100–120 Hz fundamental; J. Michael Tatum’s English dub is slightly warmer at 105–125 Hz. Women shifting for Erwin typically need −8 to −10 semitones combined with chest formant targeting.
Can I use an Erwin Smith voice mod in Discord without a kernel driver?
Yes. VoxBooster routes audio entirely through the Windows WASAPI API with no kernel driver, so it is safe alongside anti-cheat systems. In Discord, simply select the VoxBooster virtual microphone as your input device in Voice & Video settings.
How much clean audio do I need to train an Erwin AI voice model?
A usable model requires 15–30 minutes of clean isolated speech — no background music or sound effects. AOT OST tracks bleed into many scene recordings, so sourcing isolated dub recordings or clean audio rips is important. More data covering both Erwin’s measured calm and full rally-cry intensity produces a more versatile model.
Is cloning Erwin’s voice legal for personal streaming and Discord use?
For non-commercial fan use — streaming, gaming, Discord roleplays — enforcement against fictional character voice impressions is rare. For any commercial project, monetized content, or products, review Wit Studio, MAPPA, and Funimation/Crunchyroll character licensing guidelines before publishing.
What is the difference between training drills and DSP settings for voice impression?
DSP settings (pitch shift, compression, EQ) apply electronic transformations to your voice in software. Training drills are physical vocal exercises that reshape your natural resonance — ribcage breathing, vowel elongation, sustained projection practice. The best results combine both: drills bring your natural voice closer to the target, DSP covers the remaining gap.
Does AI voice cloning require a GPU for real-time use?
For real-time AI voice conversion, a GPU (GTX 1060 or better) reduces latency to sub-300ms, which is the practical threshold for live use. CPU-only inference adds 500–800 ms, making it viable only with push-to-talk discipline. Text-to-speech generation for clips and voiceovers runs fine on CPU since real-time playback is not required.
Mastering Erwin Smith’s voice is as much a performance craft as a technical exercise. The DSP settings give you the frequency foundation; the training drills give you the physical technique that makes the impression feel inhabited rather than processed. For the full vocal character — the micro-expressions in Ono’s delivery, the specific chest resonance in Tatum’s performance — AI voice cloning closes the final gap that no parameter can replicate. If you want to go beyond single-character impressions, the anime voice changer guide covers the broader workflow, and the epic narrator voice tutorial shares relevant techniques for building commanding, authoritative vocal presences from scratch.
Start the free trial of VoxBooster — Windows 10/11, no kernel driver, sub-300ms AI cloning, WASAPI routing. Free for 3 days, then from $6.99/month.