Art of Novel Signals: Predicting and Forecasting with High Confidence - SBIR Topic DPA26BZ04-DV015
Funding Amount:
$2,000,000
Deadline to Apply:
August 19th, 2026
Objective:
To develop and demonstrate a predictive/forecasting model that leverages Temporal Knowledge Graph Forecasting using In-Context Learning from novel, multilingual, and multimodal data. The goal is to develop and test a capability that increases the forecasting timeline from days to weeks in advance of an event, while increasing forecasting precision to at least 90%.
Description:
The frontier of predictive AI is running into a data wall: the high-quality open text that has powered recent models is largely exhausted, and the two ways around it both have ceilings. Reusing existing data yields little after a few passes, and synthetic data degrades models once it grows past a curated minority of the mix. Neither route creates a genuinely new signal.
This SBIR idea seeks to break the wall with signal that was never in the training distribution to begin with: multilingual radio, local news, and community reporting from data-sparse regions. This is a large, almost entirely untapped reservoir of high-value, time-sensitive information that the open web never captured and that synthetic generation cannot manufacture. The central bet of this effort is that conflict and instability signal surfaced from this audio, fed into a temporal-knowledge-graph forecasting model, materially improves geopolitical forecasting precision and warning time in exactly the environments where conventional collection is sparse or denied.
The approach has four interlocking parts:
A radio data engine.
Per-language automatic speech recognition (ASR) for predominantly oral languages.
A synthetic-data strategy that could make broad language coverage affordable.
The forecasting integration that turns transcribed audio into an early warning.
Data
Published scaling work on low-resource ASR (Akera et al., 2025) shows that fine-tuning Whisper Large-v3 (state-of-the-art automatic speech recognition model trained on over 5 million hours of labeled data) reaches usable accuracy at roughly 50 hours of transcribed audio per language and crosses the 10% word-error-rate threshold near 200 hours, with gains flattening beyond that.
For this effort, two adjustments are core areas of importance.
First, those results that were obtained on clean, single-speaker audio; radio is noisy and multi-speaker, so one should expect higher error rates on raw signal and budget for data curation, not just volume.
Second, the 200-hour figure is per language, and the commitment is to be operationally relevant, predominantly oral languages, which are precisely the hardest cases.
The collection uses two complementary methods.
Where stations stream online, the performer will ingest them directly, which extends the reach far beyond the range of any single receiver.
In the low-connectivity environments this program targets, however, many stations never reach the internet at all, and in-region software-defined radio receivers are the only way to capture them.
This is precisely why the physical radio listeners matter.
The audience is not a fallback, but rather a sole means of reaching signal that no online source can carry.
When combined, these two methods make coverage independent of both connectivity and device placement.
The performer will further supplement this audio with local news feeds and community audio from social-media channels (Telegram, WhatsApp, etc.) in the target countries.
Speech is gated to the roughly 15% of transmissions that contain it, draft transcripts are bootstrapped with Whisper Large-v3, and native-speaker annotators (correcting function).
Annotators could be sourced at reasonable costs (around $5/hour) through the performer’s existing in-region network and from diaspora communities; where channels carry human captions, found data further lowers the associated cost.
Proposed benchmark
A two-dimensional benchmark that pairs word-error-rate with hours of training data, reported per language and tagged by acoustic condition, plus a third axis comparing all-real data against real-plus-synthetic mixes.
The headline metric becomes the real transcription hours saved to reach a fixed error rate, which is simultaneously a clean scientific result and a direct cost argument.
Metrics
Across the regions of interest (Central and Southeast Asia, East and Northeast Africa, and South America), the current SOA forecasting precision is approximately 80%, and the novel-signal approach is expected to raise this toward 90%.
Precision alone is an incomplete measure; a model could score high on precision while still missing many true events, so a strong precision figure could overstate the coverage.
As the volume of radio-derived signal grows, the system would gain a more complete picture of the event space, which makes accuracy measurable.
The program therefore begins accuracy measurement once sufficient data has been accumulated and improves against that baseline.
PHASE I:
This topic is soliciting Direct to Phase II (DP2) proposals only.
Performers can bypass Phase I by providing documentation that they have developed a theory for identifying critical knowledge and a framework for testing the theory by determining knowledge requirements and assessing technologies that facilitate multi-modal data ingestion and analyses in complex, real-world settings.
Performers should plan to operationalize and evaluate their frameworks during Phase II.
Proposals will be considered for DP2 funding based on the ability of the proposing team to build a theoretical framework-based forecasting system that leverages novel multilingual, multimodal open-source signal with in-context learning over temporal knowledge graphs to anticipate events in data-sparse environments.
Proposals must clearly demonstrate that the proposed theoretical framework and technology can satisfy the following feasibility criteria:
A working in-context-learning forecasting pipeline that operates over temporal knowledge graphs without retraining graph embeddings.
Forecast precision at the baseline (days-long) horizon at or above 80% fidelity (metrics validated by an independent team/organization) event-type data such as Armed Conflict Location and Event Data (ACLED) style political and conflict events.
Demonstrate ingestion and normalization of multilingual, multimodal open-source signal, including radio and other audio from data-sparse environments.
Demonstrate that the approach generalizes to events and regions not explicitly represented in training.
Deployment and evaluation in following strategic regions of interest:
Central and Southeast Asia
East and Northeast Africa
South America
PHASE II:
Phase II fixed milestones for this program should include a Base Period where the performers are expected to produce precision across target regions with a precision of approximately 80%, and the novel-signal approach that would raise this precision level up to 90%, while reducing within-country false-positive rates by roughly half.
Precision alone is an incomplete measure.
However, the model should be able to score high on precision while still missing a large share of true events, so a strong precision figure could overstate real-world coverage.
What this effort really wants to measure is accuracy, but this requires a complete picture of the landscape of the conflict in a region.
As the volume of signal drawn from novel community sources grows, the system gains a more complete picture of the event space, making it possible to measure accuracy and not precision alone.
The base period therefore begins accuracy measurement once sufficient data has accumulated, establishing a baseline that later milestones improve against.
The milestones should include:
Month 3
Milestone:
Stand up radio collection and ingestion for the first languages; begin bootstrap-and-correct annotation; produce initial fine-tuned ASR models and the first word-error-rate-versus-hours curve; establish the forecasting baseline at the current across-region precision.
Month 6
Milestone:
Expand language coverage; begin the real-to-synthetic ratio sweep; demonstrate a 10-day forecasting horizon; reduce within-country false-positive rates by half; begin accuracy measurement to establish a baseline; report ASR ablations isolating the contribution of the radio signal.
Month 9
Milestone:
Demonstrate the two-week horizon; advance across-region precision toward the 90% target; deliver the real-versus-synthetic benchmark; quantify robustness to noise, missing data, and language-coverage gaps.
Month 12
Milestone:
End-to-end demonstration on a live or recent real-world scenario; run the with-versus-without-radio forecasting ablation; show a 30% improvement in accuracy over the Month 6 baseline while holding precision at the 90% target; deliver a mid-program report.
Month 15
Milestone:
Extend the audio model beyond transcription to paralinguistic signal.
Extract prosodic and affective features from broadcast audio, including pitch, energy, speaking rate, and indicators of emotional arousal, valence, and agitation; stand up this feature pipeline across the target languages; and establish a forecasting baseline that combines these features with the textual event stream.
Month 18
Milestone:
Test, by ablation, whether paralinguistic features add early-warning signal beyond the textual events, that is, whether rising fear, anger, or agitation on the air anticipates events the words alone do not; deliver the final base-period report and benchmark suite covering ASR, synthetic stretch, textual-event forecasting, and the paralinguistic contribution.
Option Period (6 Months, $500k)
The primary objective of the option period is to convert the demonstrated capability into operational and contingency value.
Potential directions include:
Operational pilot and transition
Run a sustained, live forecasting feed for a single theater alongside an operational user, measuring real-world warning value over the six months and produce a transition package.
Rapid language onboarding
Demonstrate standing up a new crisis language in weeks rather than the full collection cycle, using cross-lingual transfer and synthetic augmentation, to prove a surge capability for contingencies.
PHASE III DUAL USE APPLICATIONS:
The end goal of the Phase II effort is to demonstrate a commercially deployable, validated early-warning forecasting capability built on multilingual signal sourced from data-sparse environments, extending reliable prediction from five days to a two-week horizon at higher precision.
Phase III will be oriented towards transition within DoW/Military ecosystem and further commercialization of the technology.
Funding for Phase III is obtained from the private sector or a non-SBIR/STTR Government source.
This is to develop the prototype technology into a viable product or service for sale (e.g., a deployable, ruggedized, user-friendly device) in military and intelligence community or private sector markets.
The following are the potential commercial and DoW/Military applications and use cases:
DoW/Military and IC: Indications and warning at the Combatant Command level providing and/or assisting in:
Force protection for forward-deployed forces.
Embassy and diplomatic personnel.
Operational continuity in austere or denied environments.
Support to information-environment assessment.
Humanitarian assistance and disaster-response planning.
Commercial:
Country and political-risk monitoring for multi-national operators in frontier markets (mining, energy, infrastructure).
Operational-continuity and supply-chain risk for banks and insurers.
Early warning for Non-Governmental Organizations and humanitarian operations.
Who will win?
If you can achieve the objective above better than any other company on the market, you have a very high-likelihood of success and should apply.
Who is eligible to apply?
Any company that meets the following criteria:
For-profit company
U.S.-owned and controlled.
500 or fewer employees (including affiliates)
How Can BW&CO Help?
1) End-to-end support including, strategy, writing of the full proposal, and administrative & compliance support.
2) Proposal strategy and review.
3) Administrative & compliance support.
Request to talk with a member of our team by completing the form below: