Hearing anemia before the blood draw
A live clinical deployment at Shrimad Rajchandra Medical and Research Institute in Gujarat, applying Sona-2 to non-invasive anemia screening at roughly 100 assessments per day — in Gujarati and Hindi, against concurrent CBC hemoglobin.
- Partner
- SRMD Institute · Gujarat
- Condition
- Anemia screening
- Phase
- 40-day hospital pilot · active
- Ground truth
- CBC hemoglobin (g/dL)
1.6 billion people. One diagnostic tool.
Anemia is the most prevalent nutritional disorder on earth. The only way to confirm it is a blood draw. In rural, high-volume clinical settings, that bottleneck is the entire problem.
Across rural Gujarat — and much of low-resource India — anemia prevalence reaches 60 to 70% in some regions. It affects maternal health, child development, and workforce productivity at population scale, and it is almost entirely preventable with early identification and iron supplementation.
The diagnostic gap isn't a knowledge problem. Every frontline health worker knows anemia is likely. The problem is confirmation. Without a blood draw you cannot objectively diagnose it, and where lab capacity is scarce and patient volume is high, that requirement means cases are missed, under-treated and under-counted.
Amplifier's acoustic model detects physiological correlates of anemia — tissue oxygen deficit, altered respiratory drive, changes in phonation energy — from 20 seconds of prompted speech. No consumables. No needles. Any smartphone.
No other biomarker for anemia exists at population scale. You must do a blood draw. That is exactly what we are trying to change.
Three forces. One clinical deployment.
A population health crisis, a hospital with the patient volume and infrastructure, and an acoustic model with proven signal direction.
SRMD Institute · Gujarat
Active anemia outreach at ~100 assessments per day. Captures voice sessions alongside the daily CBC workflow and returns hemoglobin labels weekly.
Amplifier Health
Built the Sona-2 Large Acoustic Model and the India web application. Driving local fine-tuning and calibration to Hindi and Gujarati populations.
Label exchange
Session ID matched to hemoglobin from a concurrent CBC draw, weekly. 300 linked sessions is the inflection point for blinded validation and model lock.
Real patients. Real ground truth.
- Population
- Rural Gujarat outpatient
- Daily volume
- ~100 assessments/day
- Sample type
- B2C app · live capture
- Ground truth
- CBC hemoglobin g/dL
- Languages
- Gujarati · Hindi
- Signal direction
- Confirmed
- Fine-tuning
- Active · weekly
- Model
- Sona-2, India fine-tuned
45 of 60 patients confirmed anemic by WHO criteria
The model flagged elevated risk at a rate consistent with confirmed CBC anemia, and prevalence matched regional epidemiological data. This phase established signal direction and started the fine-tuning loop — it is not a report of clinical performance. Sensitivity and specificity will be established at the 300-session blinded evaluation threshold.
Screen. Refer. Treat.
One pathway from first contact to confirmed intervention — with voice replacing the blood draw at the triage layer.
Screen — frontline identification
Current triage depends on symptom recognition and clinical suspicion, neither of which reliably catches early-stage anemia. Sona-2 adds a passive acoustic layer at first contact.
Refer — risk stratification
A three-tier output — Elevated Risk, Monitor, Clear — lets frontline workers prioritize which patients need confirmatory lab work today. Where lab capacity is limited, that stratification is operationally decisive.
Treat — confirmation and intervention
Elevated Risk patients are referred for confirmatory hemoglobin measurement. Confirmed diagnosis triggers iron supplementation, dietary counseling, or escalation.
How Sona-2 reads anemia in voice
Five stages from raw audio to clinically actionable triage output. No blood required at any point in the pipeline.
Audio capture
Any recording device — smartphone, tablet, basic handset. Patients respond to structured prompts in Gujarati or Hindi, generating roughly 20 seconds of continuous speech.
Feature extraction
Audio is decomposed into 88 acoustic features using the extended Geneva Minimalistic Acoustic Parameter Set, capturing correlates of tissue oxygen deficit, fatigue and altered respiratory drive.
Encoder
Features pass through a contrastive language-audio pretraining encoder backed by a hierarchical token-semantic audio transformer, mapping them into the model's learned embedding space.
Sona-2 inference
The Large Acoustic Model runs inference against its trained anemia phenotype, fine-tuned on local patient data and calibrated for this population, clinical environment and language pair.
Three-tier triage output
Elevated Risk, Monitor, or Clear — each with a confidence score and feature-level explainability, designed to tell the health worker which patients need a blood draw today.
One playbook. Every high-prevalence geography.
The barrier to deployment is a smartphone and 20 seconds of speech. That asymmetry compounds with each new language and geography.
| Dimension | Figure | Context |
|---|---|---|
| Affected globally | 1.6B | Most prevalent nutritional deficiency on earth |
| India prevalence | 60–70% | Some states and regions, per NFHS data |
| Grant status | Public | In progress · validation is the trigger |
| Post-validation target | 6M | Assessments per year with the existing partner |
| Language expansion | Any | Same pipeline, new fine-tune — not a new build |
Three milestones. One clear path.
300-session calibration dataset locked
Complete the linked dataset of voice sessions with matched hemoglobin values from concurrent CBC draws, then enter formal blinded evaluation against a pre-specified threshold.
Sona-2 anemia model locked for India
Once blinded evaluation clears, the fine-tuned model is locked and the evidence package assembled: sensitivity, specificity, confidence calibration, prevalence context.
Six million assessments annually
A validated model opens the path to scaled deployment with the existing clinical partner — and the same architecture becomes the template for the next high-prevalence geography.
Working in high-prevalence anemia populations?
If your organization runs frontline health worker programs or funds population-level screening, the same pipeline, fine-tuning loop and label exchange protocol are replicable to your population and language.
All performance data reflects pilot-phase directional findings only. Sensitivity and specificity will be established at the 300-session blinded evaluation threshold. Wellness and screening signals, not diagnoses.