Jivetalking 🕺

Professional podcast audio preprocessing - broadcast-quality results with zero audio engineering knowledge required

---

What It Does

Raw microphone recordings are messy: room rumble, background hiss, awkward silences, inconsistent volume, harsh sibilance. Jivetalking fixes all of this automatically, transforming raw voice recordings into broadcast-ready audio at -18 LUFS (the broadcast/podcast standard).

jivetalking presenter1.flac presenter2.flac presenter3.flac

That's it. No configuration, no knobs to tweak, no audio knowledge required.

---

Design Philosophy

Principle	Implementation
Best outcome by default	Professional results with zero configuration
Quality over speed	Four-pass processing for measurement-driven accuracy
Transparency over depth	Every filter prioritises natural sound
Adaptive everything	Parameters tune automatically to your specific recording

---

The Filter Chain

Jivetalking's processing pipeline draws inspiration from legendary studio hardware: the Drawmer DS201 noise gate, noise suppressor, Teletronix LA-2A optical compressor, and UREI 1176 limiter. Each filter in the chain prepares the audio for the next.

Pass 1: Analysis

Measures your audio's characteristics to drive adaptive processing:

• Integrated loudness, true peak, loudness range (EBU R128)
• Noise floor and silence profile
• Spectral characteristics (centroid, rolloff, kurtosis, skewness)
• Dynamic range and transient sharpness

Pass 2: Adaptive Processing

Filter	Inspiration	What It Does
High-pass	DS201 side-chain	Removes subsonic rumble (50–60 Hz, adaptive to voice)
Low-pass	DS201 side-chain	Removes ultrasonic content that triggers false processing
Noise reduction	Non-Local Means	Adaptive Non-Local Means (anlmdn) denoiser with compand residual suppression
Gate	DS201 expander	Soft expansion for natural inter-phrase cleanup
Declicker	DC-1	Autoregressive (AR) interpolation click/pop remover
Compressor	LA-2A	Programme-dependent optical compression with ~10ms attack
De-esser	-	Tames sibilance (adaptive intensity based on spectral rolloff)

Pass 3 & 4: Loudness Normalisation

Two-stage EBU R128 loudness normalisation using FFmpeg's loudnorm filter:

Pass	What It Does
Pass 3: Measure	Analyses processed audio to get integrated loudness, true peak, LRA, and threshold
Pass 4: Normalise	Applies loudnorm with linear mode using Pass 3 measurements; UREI 1176-inspired peak limiter creates headroom for full linear gain

Why This Order Matters

Each filter prepares audio for the next:

1. Rumble removal before spectral analysis - prevents low-frequency artifacts from confusing noise profiling
2. Denoising before gating - lowers the noise floor so the gate threshold can be set optimally
3. Gating before compression - removes silence before dynamics processing amplifies room tone
4. Compression before de-essing - compression emphasises sibilance; de-essing corrects it
5. Normalisation last - sees the fully processed signal for accurate loudness targeting
6. Limiter before loudnorm - creates headroom so loudnorm can apply full linear gain without clipping or falling back to dynamic mode

Why Adaptive Matters

A dark-voiced narrator doesn't need aggressive de-essing. Pre-compressed audio doesn't need heavy compression. Clean studio recordings need different gating than noisy home offices.

Jivetalking measures your specific audio and adapts every filter automatically. The DS201-inspired gate tunes its threshold to your measured noise floor. The noise reducer adapts its compand expansion based on your measured noise profile. The LA-2A-inspired compressor adjusts ratio and release based on your dynamic range.

---

Workflow

┌─────────────────────────────────────────────────────────────┐
│  1. Record                                                  │
│     Each presenter records individually → export as FLAC    │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  2. Process                                                 │
│     $ jivetalking *.flac                                    │
│     Output: *-processed.flac (level-matched at -18 LUFS)    │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  3. Edit in Audacity                                        │
│     • Import all processed files                            │
│     • Top/tail and remove flubs                             │
│     • Select all tracks → Tracks menu → Mix → Mix to Mono   │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  4. Finalize                                                │
│     • Analyze → Loudness Normalization (preview to check)   │
│     • Normalize to -16 LUFS (dual-mono required)            │
│     • Export as final podcast file                          │
└─────────────────────────────────────────────────────────────┘

---

Installation

Pre-built Binaries

Single binary with embedded FFmpeg—no external dependencies.

Download the latest release for your platform from the releases page:

# Linux (amd64)
chmod +x jivetalking-linux-amd64
sudo mv jivetalking-linux-amd64 /usr/local/bin/jivetalking

# Linux (arm64)
chmod +x jivetalking-linux-arm64
sudo mv jivetalking-linux-arm64 /usr/local/bin/jivetalking

# macOS (Intel)
chmod +x jivetalking-darwin-amd64
sudo mv jivetalking-darwin-amd64 /usr/local/bin/jivetalking

# macOS (Apple Silicon)
chmod +x jivetalking-darwin-arm64
sudo mv jivetalking-darwin-arm64 /usr/local/bin/jivetalking

Build from Source

See Development section below.

---

CLI Usage

jivetalking [flags] <files>

Flags

Flag	Description
-v, --version	Show version information
-d, --debug	Enable debug logging to jivetalking-debug.log
--logs	Save detailed analysis reports

Examples

# Process multiple files
jivetalking presenter1.flac presenter2.flac

# Process with debug logging and analysis reports
jivetalking -d --logs recording.flac

# Process all FLAC files in directory
jivetalking *.flac

Output

Processed files are saved in the same directory with -processed suffix:

• recording.flac → recording-processed.flac

---

Development

# Clone with submodules (ffmpeg-statigo provides embedded FFmpeg)
git clone --recursive https://github.com/linuxmatters/jivetalking
cd jivetalking

# Or initialise submodules after cloning
just setup

# Build
just build

# Run tests
just test

Project Structure

cmd/jivetalking/main.go     # CLI entry point (Kong + Bubbletea)
internal/
├── audio/                  # FFmpeg demuxer/decoder wrapper
├── processor/
│   ├── analyzer.go         # Pass 1: ebur128 + astats + spectral analysis
│   ├── processor.go        # Pass 2: adaptive filter chain execution
│   ├── filters.go          # Filter chain configuration and building
│   └── adaptive.go         # Measurement-driven parameter tuning
└── ui/                     # Bubbletea TUI model and views

Design Documentation

• Gate: Drawmer DS201 - Soft expander gate with adaptive threshold
• Declick: CEDAR DC-1 - Autoregressive declicker
• Compressor: LA-2A - Programme-dependent optical compression
• Limiter: UREI 1176 - Adaptive peak limiter
• Spectral Analysis - How measurements drive adaptive processing

---

Contributing

# Run tests before committing
just test

# Create a release (validates format, creates annotated tag)
just release X.Y.Z

# Preview changelog
just changelog

• Follow Conventional Commits format
• Use just build for releases (CGO + version injection required)
• See AGENTS.md for full development guidelines