CalibraSpeakingPitch

Natural speaking pitch detection for voice profiling. Detects the median fundamental frequency of a person's voice when speaking naturally.

Quick Start

Kotlin

// From audio samples (16kHz mono)
val speakingPitch = CalibraSpeakingPitch.detectFromAudio(audioSamples)

if (speakingPitch > 0) {
    println("Speaking pitch: $speakingPitch Hz")
    val note = CalibraMusic.hzToNoteLabel(speakingPitch)
    println("Closest note: $note")
}

// Or from existing pitch contour
val contour = pitchExtractor.extract(audio, 16000)
val speakingPitch = CalibraSpeakingPitch.detectFromPitch(contour.toPitchesArray())

Swift

// From audio samples (16kHz mono)
let speakingPitch = CalibraSpeakingPitch.detectFromAudio(audioMono: audioSamples)

if speakingPitch > 0 {
    print("Speaking pitch: \(speakingPitch) Hz")
    let note = CalibraMusic.hzToNoteLabel(speakingPitch)
    print("Closest note: \(note)")
}

// Or from existing pitch contour
let contour = pitchExtractor.extract(audio: audio, sampleRate: 16000)
let speakingPitch = CalibraSpeakingPitch.detectFromPitch(pitchesHz: contour.toPitchesArray())

When to Use

Scenario	Use This?	Why
Detect natural voice pitch	Yes	Core use case
Classify voice type	Yes	Based on frequency range
Suggest shruti/tonic	Yes	Recommend musical reference note
Voice health tracking	Yes	Monitor changes over time
Detect singing range	No	Use CalibraVocalRange
Real-time pitch display	No	Use CalibraPitch

Methods

detectFromAudio

Detect natural speaking pitch from audio samples. Analyzes spoken audio to find the median fundamental frequency. Works best with natural speech samples. Automatically resamples to 16kHz internally if needed.

Kotlin

fun detectFromAudio(audioMono: FloatArray, sampleRate: Int = 16000): Float

Swift

static func detectFromAudio(audioMono: [Float], sampleRate: Int = 16000) -> Float

Parameters

Parameter	Type	Default	Description
audioMono	FloatArray / [Float]	--	Mono audio samples (normalized -1 to 1)
sampleRate	Int	16000	Sample rate of the input audio in Hz

Return Value

Detected frequency in Hz, or -1 if detection failed (not enough voiced audio).

Example

// Kotlin
val audioData = SonixDecoder.decode("/path/to/speech.mp3")
val samples16k = SonixResampler.resample(audioData.samples, audioData.sampleRate, 16000)
val speakingPitch = CalibraSpeakingPitch.detectFromAudio(samples16k)

// Swift
let audioData = SonixDecoder.decode(path: "/path/to/speech.mp3")!
let samples16k = SonixResampler.resample(samples: audioData.samples,
                                         fromRate: Int(audioData.sampleRate),
                                         toRate: 16000)
let speakingPitch = CalibraSpeakingPitch.detectFromAudio(audioMono: samples16k)

detectFromPitch

Detect natural speaking pitch from an existing pitch contour. Analyzes pitch values to find the median pitch. Useful when you have already extracted pitches from audio.

Kotlin

fun detectFromPitch(pitchesHz: FloatArray): Float

Swift

static func detectFromPitch(pitchesHz: [Float]) -> Float

Parameters

Parameter	Type	Description
pitchesHz	FloatArray / [Float]	Pitch values in Hz (-1 for unvoiced frames)

Return Value

Detected frequency in Hz, or -1 if detection failed.

Example

// Kotlin
val extractor = CalibraPitch.createContourExtractor()
val contour = extractor.extract(audioSamples, sampleRate = 16000)
val speakingPitch = CalibraSpeakingPitch.detectFromPitch(contour.toPitchesArray())
extractor.release()

// Swift
let extractor = CalibraPitch.createContourExtractor()
let contour = extractor.extract(audio: audioSamples, sampleRate: 16000)
let speakingPitch = CalibraSpeakingPitch.detectFromPitch(pitchesHz: contour.toPitchesArray())
extractor.release()

Typical Speaking Pitches

Voice Type	Typical Range
Bass	85--155 Hz
Baritone	110--165 Hz
Tenor	130--200 Hz
Alto	175--255 Hz
Soprano	220--330 Hz

Common Patterns

Voice Profile Setup

class VoiceProfileViewModel : ViewModel() {
    fun detectSpeakingPitch(audioPath: String) {
        viewModelScope.launch {
            val audioData = SonixDecoder.decode(audioPath) ?: return@launch
            val samples16k = SonixResampler.resample(
                audioData.samples, audioData.sampleRate, 16000
            )

            val pitchHz = CalibraSpeakingPitch.detectFromAudio(samples16k)
            if (pitchHz > 0) {
                val noteLabel = CalibraMusic.hzToNoteLabel(pitchHz)
                println("Speaking pitch: $noteLabel ($pitchHz Hz)")
            } else {
                println("Could not detect speaking pitch")
            }
        }
    }
}

Reusing an Existing Pitch Contour

If you have already extracted a pitch contour (e.g., for visualization or scoring), you can pass it directly to avoid re-processing the audio:

val extractor = CalibraPitch.createContourExtractor()
val contour = extractor.extract(audioSamples, sampleRate = 16000)

// Use contour for visualization...
updatePitchDisplay(contour)

// Also derive speaking pitch from the same contour
val speakingPitch = CalibraSpeakingPitch.detectFromPitch(contour.toPitchesArray())
extractor.release()

Platform Notes

• iOS/Android: Accepts any sample rate; internally resamples to 16kHz if needed.
• Uses median-based detection in MIDI space for robustness against outliers.
• Returns -1 if detection fails (not enough voiced audio).
• Requires several seconds of natural speech for reliable results.

Common Pitfalls

1. Singing instead of speaking -- This detects speaking pitch, not singing range. For singing, use CalibraVocalRange.
2. Not enough audio -- Short clips may not contain enough voiced frames. Aim for several seconds of natural speech.
3. Background noise -- High noise levels can affect detection accuracy. Record in a reasonably quiet environment.

Next Steps

• CalibraVocalRange -- Detect full singing range
• CalibraPitch -- Real-time pitch detection
• CalibraVAD -- Voice activity detection