TesseraSession

Streaming multi-metric session — feed audio incrementally, get results on demand.

What is a Tessera Session?

The streaming counterpart to Tessera.analyze. Instead of passing a complete PitchContour, you feed raw audio or pre-extracted pitch incrementally via addAudio / addPitch. When you want results, call getResult — breath and agility are computed in batch on the full accumulated contour; vocal range bounds (lower/upper) come from a streaming histogram that updates as pitch arrives, while the search vector is recomputed in batch from the accumulated contour.

When to Use

Quick Start

Kotlin

val session = TesseraSession.create()

recorder.audioBuffers.collect { buffer ->
    session.addAudio(buffer.samples, sampleRate = 48000)
}

val result = session.getResult()
println("Breath: ${result.breath?.controlScore}")
session.release()

Swift

let session = TesseraSession.create()

for await buffer in recorder.audioBuffers {
    session.addAudio(samples: buffer.samples, sampleRate: Int32(hwRate))
}

let result = session.getResult()
print("Breath: \(result.breath?.controlScore ?? 0)")
session.release()

Common Pitfalls

1. Call release() when done: Holds a pitch detector and range detector internally.

2. addAudio requires a detector: The default create() always creates one. If you only use addPitch, the internal detector is created but unused — a minor overhead, not an error.

3. getResult() can be called multiple times: Each call recomputes breath/agility on the accumulated data. Range uses the streaming histogram so it's cheap.

4. Max session duration is 10 minutes: After that, old pitch points are trimmed from the accumulator.

See also