Skip to main content

VoxaTrace

Voice AI today understands what you say: the words, the language, the text. But it's deaf to how you actually sound — pitch, timbre, emotion, vocal quality, rhythm, melody. Everything that makes a voice a voice, not just a transcript.

VoxaTrace is an on-device SDK that makes any application acoustically intelligent.

Eight years of R&D. Five million users in production. All running natively on Android and iOS, without a single server call.

What Speech AI Misses​

Speech AIVoxaTrace
"The user said 'hello'""The user sang A4 at 440 Hz with 92% confidence"
Words and languagePitch, melody, rhythm, vocal quality
TranscriptionAcoustic analysis
Cloud-dependentOn-device, real-time

What You Can Build​

ApplicationWhat VoxaTrace Enables
Singing appsPitch detection, real-time scoring, performance feedback
Vocal trainingIntonation analysis, progress tracking, guided exercises
Music educationEar training, sight-singing evaluation, pitch matching
Voice gamesPitch as input — sing to jump, hum to control
AccessibilityVoice-based input beyond speech recognition
Health & wellnessVocal health monitoring, breathing exercises

What You Get​

┌─────────────────────────────────────────────────────────────────┐
│                         VoxaTrace                               │
├──────────────────────────────┬──────────────────────────────────┤
│           Sonix              │            Calibra               │
│       (Audio Engine)         │      (Acoustic Analysis)         │
├──────────────────────────────┼──────────────────────────────────┤
│  • Multi-track playback (8)  │  • Pitch detection (YIN/SwiftF0) │
│  • Recording (M4A/MP3)       │  • Voice activity detection      │
│  • Pitch shifting ±12 semi   │  • Singing evaluation & scoring  │
│  • Tempo control 0.5x–2x     │  • Vocal range detection         │
│  • MIDI synthesis (SoundFont)│  • Audio effects chain           │
│  • Metronome with callbacks  │  • Octave error correction       │
└──────────────────────────────┴──────────────────────────────────┘

Performance​

MetricSpecification
Pitch detection latency~50ms
Frequency range50 Hz – 2000 Hz
Confidence threshold0.0 – 1.0 (recommended > 0.7)
Simultaneous tracksUp to 8 synchronized
Sample ratesAuto-resampling to 16kHz
Minimum AndroidAPI 24 (Android 7.0)
Minimum iOSiOS 14

Hello, Pitch Detection​

Kotlin​

val detector = CalibraPitch.createDetector()
val point = detector.detect(audioSamples, sampleRate = 16000)
println("${point.pitch} Hz @ ${(point.confidence * 100).toInt()}% confidence")
detector.close()

Swift​

let detector = CalibraPitch.createDetector()
let point = detector.detect(samples: audioSamples, sampleRate: 16000)
print("\(point.pitch) Hz @ \(Int(point.confidence * 100))% confidence")
detector.close()

Output:

440.0 Hz @ 92% confidence

Why VoxaTrace?​

On-device, not cloud:

  • Zero latency from network calls
  • Works offline
  • User data stays on device

Battle-tested:

  • 8 years of R&D
  • 5 million users in production
  • Same API on Android and iOS

Three-tier API:

// Tier 1: Just works
val detector = CalibraPitch.createDetector()

// Tier 2: Configurable
val detector = CalibraPitch.createDetector(
    PitchDetectorConfig.Builder()
        .algorithm(PitchAlgorithm.SWIFT_F0)
        .build()
)

// Tier 3: Full control
val config = PitchDetectorConfig.PRECISE.copy(confidenceThreshold = 0.6f)

Read more about why VoxaTrace →

Start Building​

5-Minute Quickstarts​

Complete App Recipes​

Understand the Concepts​

Module Deep Dives​

Installation​

// build.gradle.kts
dependencies {
    implementation("com.musicmuni:voxatrace:0.9.2")
}
// Package.swift
dependencies: [
    .package(url: "https://github.com/musicmuni/voxatrace", from: "0.9.2")
]
# Podfile
pod 'VoxaTrace', :podspec => 'https://raw.githubusercontent.com/musicmuni/voxatrace/main/VoxaTrace.podspec'

Full installation guide →

Support​