Callyope vs Ellipsis Health
Both listen to a patient's voice for clinical signal rather than for words, which is the most technically ambitious thing anyone in behavioral health AI is attempting. Callyope is building an audio and language foundation model trained specifically on psychiatric data rather than adapting a general model, which is a research position and the only one like it in this index. Ellipsis has already productised the idea, with vocal biomarker technology originating in depression and anxiety detection research and now sold as a care management agent to health plans and risk bearing providers. The gap is maturity against method. Ellipsis is deployable now and its inference is doing work in production. Callyope's approach is the better founded one if the claim ever needs defending, because a model built on psychiatric speech starts from the population it is describing.
- The model is trained specifically on psychiatric audio and language rather than adapted from a general purpose speech model, which is a different technical starting point for a population whose speech carries clinical signal.
- Building a foundation model for psychiatry is a research position rather than a product wrapper, and it is the only one of its kind in this index.
- For a research or specialist psychiatric setting, a model built on the target population is more defensible than one fine tuned onto it.
- It is productised and deployed, with a care management agent sold to health plans and at risk providers rather than a model looking for a workflow.
- The vocal biomarker lineage is published research that has been carried into a commercial product, so the technical claim has a trail behind it.
- For a plan trying to reach members between visits, an agent that conducts the conversation is closer to the outcome than a model that analyses one.
Side by Side
| Axis | C Callyope |
E Ellipsis Health |
|---|---|---|
| AI Centrality | ||
| Autonomy and Oversight Model | ||
| Model and Technology Transparency | ||
| Clinical and Operational Evidence | ||
| AI Safety and PHI Stewardship | ||
| HIPAA and BAA Posture | ||
| Security Certifications and Trust Center | ||
| FDA and Regulatory Status | ||
| AI Governance and Bias Disclosure | ||
| EHR and Interoperability Depth | ||
| Deployment Model and Data Residency | ||
| Commercial Transparency | ||
| Setting and Specialty Coverage |
Related comparisons
Other published head to head assessments involving these vendors or their closest peers. The full set for this category is on the Remote Monitoring & Chronic Care page.
Vocal biomarker inference is the most consequential claim in this pair and the least externally validated: inferring depression or anxiety from voice is a clinical assertion about a person, and neither vendor publishes independent validation of that inference or performance by language, accent, age or medication status, all of which alter voice. Speech changes caused by antipsychotic and antidepressant medication are a specific confound in this population.
Establish with both whether the system is measuring a clinical state or a medication effect, and what happens to the recording, since psychiatric audio carries protections beyond the general health privacy rule in many jurisdictions.