Canonical intents
Keep semantic actions compact and parameterized instead of creating unnecessary duplicate intents.
RESEARCH / URDU AUTOMOTIVE VOICEBuild the benchmark
Build the benchmark
before building the claim.
Our research programme starts with a structured automotive ontology, human-reviewed Urdu language data, controlled audio collection and provider-neutral benchmarking.
01 / METHODA language asset that survives
A language asset that survives
provider changes.
ASR and TTS providers can change. The long-term value is the automotive intent ontology, Urdu corpus, pronunciation assets, benchmark set and OEM-neutral command contract.
Human language review
Natural Pakistani Urdu wording must be reviewed before a dataset is frozen as Gold.
Real audio
Evaluate with recorded speakers and controlled acoustic conditions, not synthetic text-only assumptions.
Held-out evaluation
Keep a test set outside prompt tuning and provider adaptation so accuracy remains meaningful.
CURRENT R&D STATUS
Urdu first. Evidence before expansion.
Urdu for Pakistan is the first automotive voice evaluation language. Sindhi, Punjabi Shahmukhi and Pashto remain roadmap languages on the same canonical architecture.
LINGUISTIC REVIEW IN PROGRESSPILOT AUDIO READY TO COLLECTPROVIDER BENCHMARK PLANNED