Keyword Spotting

aiOla’s Keyword Spotting (KWS) brings real-time, low-latency recognition of mission-critical words and phrases to every product built on the Jargonic ASR family. Whether you need to detect compliance terms, extract industry jargon from audio, or trigger actions based on specific phrases, KWS lets you listen for exactly what matters—nothing more, nothing less.

Pro Tip: Need complete transcriptions and jargon detection in the same request? Just pass a keywords dictionary—Jargonic returns clean text with your preferred spelling.

aiOla Models Behind the Scenes

Jargonic-v2 & Jargonic-v2-Flash

Our flagship multilingual ASR models:

v2 – Maximum accuracy for production workloads
v2-Flash – Ultra-low latency with a small WER trade-off

AdaKWS – Next-gen Jargon Detector

A task-specific model that boosts keyword accuracy by 6% overall (and 16% in English). AdaKWS excels at noisy environments and niche vocabularies—perfect for healthcare, legal, finance, and engineering.

Typical Use Cases

Industry	Example Keywords	Action
Healthcare	”metoprolol”, “CBC”, “STAT”	Auto-populate EMR fields, trigger medication alerts, escalate urgent orders
Finance	”SOX compliance”, “K-1”, “FOMC”	Route to compliance team, log regulatory mentions, trigger audit workflows
Automotive	”OBD two diagnostic”, “CAN bus error”, “ECU firmware”	Update service records, schedule maintenance, notify technicians
Aviation	”ETOPS approval”, “MEL deferral”, “ACARS downlink”	Update maintenance logs, process regulatory compliance, sync flight data

Quick Start with the aiOla SDK

Below is the fastest way to add keyword spotting to a batch transcription:

1 from aiola import AiolaClient
2 
3 client = AiolaClient(api_key="YOUR_API_KEY")
4 
5 # Spoken form → canonical jargon spelling
6 keywords = {
7     "hemoglobin a one c":       "HbA1c",          # Healthcare
8     "sarbanes oxley":           "SOX Compliance", # Finance / Regulatory
9     "infrastructure as code":   "IaC"             # DevOps / Cloud
10 }
11 
12 transcript = client.stt.transcribe_file(
13     file="meeting.wav",
14     language="en",
15     keywords=keywords,          # zero-shot jargon detection
16     model="jargonic-v2"         # or "jargonic-v2-flash"
17 )
18 
19 print(transcript.text)

Best Practices

Use natural pronunciations
Write the phrase the way speakers say it ("hemoglobin a one c" instead of "HbA1c").
Limit the list
A dozen well-chosen keywords typically outperform huge lists.
Canonicalize output
Map common mispronunciations ("sarbanes oxley" → "SOX Compliance") for clean transcripts.

Technical Advantages

Zero-Shot Recognition

Unlike traditional systems requiring extensive training data, aiOla’s keyword spotting works immediately with any custom vocabulary. Simply provide your specialized terms, and the system adapts instantly.

Noise Resilience

Advanced filtering and noise cancellation algorithms ensure accurate detection even in challenging acoustic environments.

Multilingual Support

Built-in support for 120+ languages with automatic language detection, making it ideal for global applications.

Frequently Asked Questions

Does keyword spotting slow down transcription?

No—AdaKWS runs in parallel with the main ASR model.

Can I update the keyword list on the fly?

Absolutely. Pass a new keywords dictionary with every request; no retraining required.

How many languages are supported?

Over 120+. aiOla automatically detects the spoken language or you can specify it with the language parameter.

What's the maximum number of keywords I can use?

While there’s no hard limit, we recommend 10-50 keywords for optimal performance. Larger lists may impact accuracy.

Build smarter voice experiences with aiOla’s advanced keyword spotting technology.