Pitch & prosody
Pitch level, pitch variation, contour, voiced fraction.
RESEARCH ATLAS / 02
New models.
New possibilities.
We’re creating the building blocks of an intelligence that can participate in the world. Each family begins with a different question.
01 / SENSORY INTELLIGENCE
MODEL PROGRAMMEPlutoASRA new way to hear.
Speech is a living signal. The words are only one part of it.
PlutoASR is our streaming speech programme: following the words as they arrive while preserving the timing, texture, and dynamics of the voice.
The design brings transcript, speaker information, and a panel of 23 vocal dimensions into the same stream. The goal is to give an interactive system richer context for its next response.
- INPUT
- Continuous audio
- OUTPUT
- Words, speaker, vocal signals
- DESIGNED FOR
- Conversation as it unfolds
Inside the vocal measurement panel 23 dimensions / 6 groups+
Timing & rhythm
Speech rate, articulation rate, pauses, and filled pauses.
Energy & spectral detail
Loudness, variation, spectral centroid, formant dispersion, and spectral tilt.
Voice quality
Cepstral peak prominence, creak, jitter, and shimmer. Some measures need sustained vowels or clean recordings.
Turn-taking
Floor-transfer timing and interruptions.
Model-inferred signals
Arousal, valence, and dominance estimates. These are interpretations with uncertainty, not direct measurements of someone’s feelings.
Measurement and interpretation remain separate. Recording conditions affect reliability; model-inferred signals carry additional uncertainty. This panel describes the programme’s design, not a public benchmark result.
02 / REFLEX INTELLIGENCE
MODEL PROGRAMMEPrime-BaseA small model.
A decisive moment.
Which tool fits this request? Is this claim supported? Should a task move forward, wait, or ask for help?
Prime-Base explores models built around a declared set of outcomes. Language and application state go in. A structured judgment comes back, ready for software to use.
Our aim is to make these small acts of intelligence practical throughout an application, reserving deeper reasoning for the moments that need it.
Ask about Prime-BaseA JUDGMENT, DRAWN OUTILLUSTRATIVE EXAMPLE
STRUCTURED OUTCOMEroute_to_tool
Choose a tool from the application’s declared set.
{ "decision": "route_to_tool" }A concept illustration. Choices show predefined outcomes; no model is called.
03 / INTERACTION INTELLIGENCE
EXPLORATORY RESEARCHEICloser to the conversation.
A pause can change a moment. An interface should have room to notice.
Our emotional intelligence research asks how vocal signals could help machines better understand an interaction: when to wait, when to clarify, and how to respond with care.
We start with observable signals and keep inference explicit. A voice offers clues about an exchange; it does not give a machine access to a person’s inner life.
Better context. More thoughtful interaction.
Room for uncertainty.
Read our principlesRoom for uncertainty.