AUDIO AI PORTFOLIO

LUCAS FERNANDES NASCIMENTO

AI AUDIO EVALUATOR
VOICE & SPEECH QA
AUDIO ANNOTATION SME
SOUND DESIGNER

BAURU, BRAZIL / UTC-3 / 20+ HRS WEEK / FL STUDIO / 48-24 WAV

LFN / 2026 / BUILT FOR SIGNAL
01 /

Voice Demo

Four voice samples with raw and processed versions, demonstrating English technical narration, conversational delivery, expressive reading, and native Brazilian Portuguese. Recorded with Audio-Technica AT2020, M-Audio M-Track Solo, and edited/exported in FL Studio.

01 / Technical Narration EN / 48kHz / 24-bit WAV

Neutral English narration demonstrating clarity, pacing, pronunciation, and technical delivery.

Raw
Processed
02 / Conversational EN / 48kHz / 24-bit WAV

Natural English explanation style, showing spoken clarity, timing, and less formal delivery.

Raw
Processed
03 / Expressive EN / 48kHz / 24-bit WAV

Expressive read demonstrating tone control, atmosphere, and dynamic vocal presence.

Raw
Processed
04 / PT-BR Sample PT-BR / 48kHz / 24-bit WAV

Native Brazilian Portuguese sample demonstrating natural articulation, pacing, and bilingual voice capability.

Raw
Processed
02 /

Audio Restoration -- Before & After

Before/after restoration cases showing noise reduction, room cleanup, clipped vocal repair, and AI-generated audio denoising. All WAV files are short 48kHz / 24-bit exports.

01 / Noisy Speech 48kHz / 24-bit WAV
Before
After
Problem Background noise / low SNR
Process Noise reduction, cleanup, EQ balancing, level matching
Result Cleaner speech presence with reduced background noise
02 / Room Reflections 48kHz / 24-bit WAV
Before
After
Problem Excessive room reflections and distant vocal tone
Process Reflection reduction, EQ cleanup, intelligibility enhancement
Result More focused voice with reduced room buildup
03 / Clipped Vocal 48kHz / 24-bit WAV
Before
After
Problem Overloaded input / clipped vocal peaks
Process De-clip style restoration, harshness control, level balancing
Result Reduced harshness and more controlled transient behavior
04 / AI Audio Shimmer / Hiss 48kHz / 24-bit WAV
Before
After
Problem AI-generated audio with high-frequency shimmer / hiss
Process Denoising pass followed by spectral artifact check
Result Reduced high-frequency haze while preserving the main spectral structure
Spectrogram comparison showing AI-generated audio shimmer and hiss before and after denoising
Compare the frequency content before and after denoising. Shimmer appears as diffuse haze in the high frequencies; after denoising, that haze is visibly reduced while the main harmonic and transient structure remains largely preserved.
03 /

Audio Annotation & Technical Evaluation Sample

Structured annotation and technical evaluation of an unreleased human-made production excerpt, combining timestamped segmentation, instrumentation tagging, vocal-sample treatment notes, mix behavior, measured fidelity review, and evaluation-style technical annotation.

Audio Annotation & Technical Evaluation Sample -- Short Overview PDF overview / timestamped review / MP3
Section Summary
Purpose Portfolio annotation sample demonstrating timestamped segmentation, instrumentation tagging, vocal-sample treatment notes, mix behavior, measured fidelity review, and evaluation-style technical annotation.
Arrangement / Production The excerpt develops from a chopped spoken-word interview intro into a filtered 808-driven electronic beat, using gating, reverse edits, cut-and-repeat phrasing, synth pads, bell-like synth details, and LFO-driven synth-bass movement.
Technical Behavior The intro is more open and midrange-forward, while the beat sections are moderately compressed, center-focused, and low-end dominant, with the strongest measured low-frequency focus around 64.6 Hz.
Evaluation Focus The main production signature is the transformation of spoken-word material into rhythmic and musical texture. The main technical flag is peak/headroom: true peak reaches around +1.2 dBFS, so the file is not broadcast-safe without correction.

This document presents a structured audio annotation of an unreleased 2016 production excerpt, combining musical segmentation with measured technical review. Integrated loudness is estimated at -8.6 LUFS, with 9.0 LU loudness range and true peak around +1.2 dBFS. The mix is narrow-to-moderate, center-focused, and mostly mono-compatible by correlation, with controlled transients and no proven sustained clipping, codec artifacts, or phase problems. Overall fidelity is best described as clean demo / mid-tier production, with a clear note that peak/headroom correction is needed before broadcast-safe delivery.

04 /

Sound Design Reel -- Breakcore / Synth Voice Processing

Original sound design and production excerpt in a breakcore-influenced experimental electronic style. The piece combines bass design, synth/FM texture, processed voice material, fast breakbeat edits, dense rhythmic programming, and club-oriented low-end pressure.

Original Production Excerpt MP3 / breakcore-influenced experimental electronic
Source Original production excerpt by Lucas Fernandes Nascimento
Style Breakcore-influenced experimental electronic
Focus Bass design, synth/FM texture, processed voice material, fast breakbeat edits, dense rhythmic programming
Role Composition, sound design, arrangement, processing, mixing, mastering
File sound-design-reel-breakcore-synth-voice.mp3
  • Bass design, synth/FM texture, processed voice material, filtered movement, and atmospheric buildup.
  • Breakcore-influenced drum programming, fast breakbeat edits, dense rhythmic movement, bass pressure, and high-energy arrangement.
05 /

Evaluation Rubrics

Framework for Evaluating Audio Quality in Generative Models PDF preview / rubric discipline

Evaluating generative audio is different from evaluating finished records. You're not judging a final product -- you're judging a model's attempt to produce one. This framework is the working discipline I apply across audio evaluation and annotation workflows, holding two frames at once: is this good audio, and is this a good output given what was asked.

Dimensions 1. Audio Quality (technical)
2. Mix Quality (engineering)
3. Musicality (perceptual/aesthetic)
4. Prompt Adherence (generative-specific)
5. Vocal Quality (when present)
Featured Intent vs. Failure

The core discipline is distinguishing stylistic choices from technical failures. Four principles I apply:

  • Genre baseline first -- establish what the genre accepts as normal before flagging
  • Consistency test -- throughout-with-intention is stylistic; once-and-breaks-continuity is failure
  • Engineer test -- would a professional release this take?
  • Ambiguous cases -- autotune, distortion, lo-fi, pitch drift, dry vocals, compressed dynamics all shift stylistic/failure based on genre context
06 /

Selected Releases

Selected Releases Curated external release links

Curated releases showing electronic production, remix work, sound design, vocal manipulation, beat programming, and international label context.

Year Artist Title Label / Context Link
2020 Casinhas Fofas “Raio” -- Casinhas Fofas Remix MAMBA REC, Brazil PLAY ->
2019 AKAAKA “HOPEBACK” XXIII, Portugal PLAY ->
2019 AKAAKA “Wacky” Bruk Broken Beats, Brazil PLAY ->
2016 AKAAKA “The Sound” THUMP / Zambi, USA-Brazil PLAY ->
2016 AKAAKA “Dolphin” THUMP / FLUXXX, USA-Brazil PLAY ->
2014 AKAAKA “Benzydamine” LKHP, Germany PLAY ->
2013 AKAAKA “São Sebastião” Unlabelled / self-release PLAY ->
2011 AKAAKA “You Make You Feel” -- Dumbo Gets Mad Remix Bad Panda Records PLAY ->
07 /

Tools & Setup

DAWs / Editing

  • FL Studio
  • Ableton Live
  • Audacity
  • Cubase

Recording

  • Audio-Technica AT2020
  • M-Audio M-Track Solo
  • 48kHz / 24-bit WAV
  • Windows 10+

Restoration / Review

  • FL Studio editing/export
  • Denoising/restoration workflow
  • Spectrogram-based artifact review
  • Loudness/headroom review

AI / Data Workflow

  • Rubric-based QA
  • A/B model evaluation
  • Human-made music dataset annotation
  • Generative audio model evaluation
  • Timestamped segmentation
  • Artifact and vocal-processing detection

Downloads

Contact