Cloud VoiceMark - VoiceMark

The Cloud VoiceMark is a per-utterance acoustic portrait. It captures the word-level f0, energy, and spectral shape of one specific recording. It changes every time the same person speaks.

What it represents

A live signal that this person spoke right now
Useful as a visual liveness confirmation in your UI
Detects synthetic or cloned audio — TTS shows flat f0 and uniform spectral bands

Where it comes from

The Cloud VoiceMark data is returned by /v1/verify inside cloud_data.words:

{
  "cloud_data": {
    "words": [
      { "word": "Wild", "f0": 142.3, "rms": 0.08, "bands": [0.4, 0.6, 0.3, 0.2] },
      { "word": "roses", "f0": 128.1, "rms": 0.11, "bands": [0.5, 0.7, 0.4, 0.3] },
      ...
    ],
    "glyph_params": { ... }
  }
}

Word-level fields

Field	Type	Description
`word`	string	Transcribed word
`f0`	number	Fundamental frequency (Hz). Real speech: 80–300 Hz. TTS: often flat or zero.
`rms`	number	Root mean square energy (0–1). Indicates emphasis and projection.
`bands`	number[4]	Spectral energy in four sub-bands. Use to colour-map word clouds.

Rendering guidance

Use f0 to size or weight each word in a phrase cloud — louder, more stressed words render larger. Map bands[0] (low) → warm colours, bands[3] (high) → cool colours. The variation across words signals authentic human speech.

The liveness_score in the verify response encodes the f0 variation analysis. If you only need the score and not the raw data, use that field.

⌘I