On-device SDK making apps acoustically intelligent: real-time pitch detection, singing evaluation, vocal quality and melody analysis, multi-track audio engine, low-latency offline operation.
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.
| Speech AI | VoxaTrace |
|---|---|
| "The user said 'hello'" | "The user sang A4 at 440 Hz with 92% confidence" |
| Words and language |
| Application | What VoxaTrace Enables |
|---|
┌──────────────┬──────────────┬──────────────┬──────────────┬────────────┐
│ Sonix │ Tona │ Tessera │ Accura │ Calibra │
│ Audio I/O │ Pitch │ Voice metrics│ Intonation │ Singing │
│ │ │ │ scoring │ eval │
├──────────────┼──────────────┼──────────────┼──────────────┼────────────┤
│ • Player │ • Detection │ • Breath │ • EQ / JI │ • LiveEval │
│ • Recorder │ • Processing │ • Agility │ per-note │ • MelodyEv │
│ • Mixer │ • Analysis │ • Range │ deviation │ • NoteEval │
│ • Encoder │ (histogram │ • Speaking │ • 0–100 │ • VAD │
│ • Decoder │ transcr.) │ pitch │ scoring │ • Effects │
│ • Metronome │ │ │ │ │
│ • MIDI synth │ │ │ │ │
└──────────────┴──────────────┴──────────────┴──────────────┴────────────┘
Common: MusicTheory (Hz/MIDI/cents conversions, shruti alignment)
dependencies {
implementation("com.musicmuni:voxatrace:2.0.0")
}
dependencies: [
.package(url: "https://github.com/musicmuni/voxatrace", from: "2.0.0")
]
pod 'VoxaTrace', :podspec => 'https://raw.githubusercontent.com/musicmuni/voxatrace/main/VoxaTrace.podspec'
VT.initializeForServer("sk_live_…") // see docs for proxy / attestation flows on mobile
val detector = PitchDetection.createDetector()
val point = detector.detect(audioSamples, sampleRate = 16000)
println("${point.pitch} Hz @ ${(point.confidence * 100).toInt()}% confidence")
detector.close()
let detector = PitchDetection.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
Commercial. See LICENSE.
| Pitch, melody, rhythm, vocal quality |
| Transcription | Acoustic analysis |
| Cloud-dependent | On-device, real-time |
| Singing apps | Pitch detection, real-time scoring, performance feedback |
| Vocal training | Intonation analysis, progress tracking, guided exercises |
| Music education | Ear training, sight-singing evaluation, pitch matching |
| Voice games | Pitch as input — sing to jump, hum to control |
| Metric | Specification |
|---|
| Pitch detection latency | ~64 ms (1024-sample window at 16 kHz, BALANCED) |
| Default frequency range | 80 Hz – 1000 Hz (configurable per VoiceType) |
| Sample rates | Auto-resampling to 16 kHz internally (ADR-017) |
| Minimum Android | API 24 (Android 7.0) |
| Minimum iOS | iOS 15 |
Surfaced from shared tags and platforms — no rankings paid for.