Skip to main content

Calibra Overview

Calibra is the audio analysis module of VoxaTrace, providing:

  • Pitch Detection — Real-time fundamental frequency detection
  • Voice Activity Detection — Detect when someone is speaking or singing
  • Vocal Range Detection — Determine a singer's comfortable range
  • Singing Evaluation — Score singing accuracy against a reference

Calibra in Action​

Real-Time Pitch Feedback​

User sings: "Hap-py birth-day to you"

Timeline:     0.0s    0.5s    1.0s    1.5s    2.0s
              ├───────┼───────┼───────┼───────┤
Reference:    C4      C4      D4      C4      F4
Detected:     C4      C4      D4      C#4     F4
              ✓       ✓       ✓       ~       ✓

Score: 4/5 notes matched = 80%

Voice Activity Detection​

Audio input:  [noise][singing~~~~~][breath][singing~~~~~][noise]
VAD output:   [  0  ][     1      ][  0   ][     1      ][  0  ]

Action:       Skip    Process      Skip    Process       Skip

Live Evaluation Flow​

                    ┌──────────────┐
    Reference ───── │              │
    Audio           │   Compare    │ ───── Score (0.0–1.0)
                    │   Pitches    │
    Student  ─────  │              │
    Audio           └──────────────┘
                          │
                          ▼
              ┌───────────────────────┐
              │ Segment 1: 85%        │
              │ Segment 2: 92%        │
              │ Segment 3: 78%        │
              │ ───────────────────── │
              │ Overall: 85%          │
              └───────────────────────┘

Key Features​

Pitch Detection​

  • Real-time F0 detection with two algorithms (YIN and SwiftF0)
  • Works with singing and speech
  • Frequency range: 50Hz – 2000Hz
  • Low latency (~50ms)
val detector = CalibraPitch.createDetector()
val point = detector.detect(samples, sampleRate = 16000)
// point.pitch = 440.0, point.confidence = 0.92

Voice Activity Detection (VAD)​

Multiple detection backends for different use cases:

BackendBest ForLatency
General (Energy)Simple apps, low power<1ms
Speech (Silero)Voice assistants, transcription~5ms
Singing (YAMNet)Karaoke, singing detection~20ms
SingingRealtimeReal-time singing apps~5ms
val vad = CalibraVAD.create(VADConfig.SINGING)
val result = vad.analyze(samples, sampleRate)
// result.isSpeaking = true, result.level = Level.MODERATE

Vocal Range Detection​

Guided range finding that detects a singer's comfortable range:

val session = VocalRangeSession.create()
session.start()
// Guide user through high and low notes
val range = session.finish()
// range.lowNote = 48 (C3), range.highNote = 72 (C5)

Singing Evaluation​

Two evaluation modes:

Singalong Mode — Evaluate while singing along with reference audio

  • Real-time pitch comparison
  • Note-by-note scoring
  • Immediate feedback

Singafter Mode — Listen to phrase, then sing it back

  • Pitch and rhythm evaluation
  • Detailed per-phrase scoring
  • Great for ear training
val session = CalibraLiveEval.create(lessonMaterial, config)
session.onSegmentComplete { result ->
    println("Segment ${result.segmentIndex}: ${result.score}%")
}
session.start()

Next Steps​

API Reference​

Guides​