What is named entity recognition (NER) in speech-to-text?

Named entity recognition is the process of finding and labeling key information in a transcript, such as names, organizations, dates, and account numbers. In voice products, NER on call transcripts combines machine learning with pattern matching to catch both free-form names and structured identifiers.

How do I use an API to highlight named entities in speech transcripts?

Set named_entity_recognition to true in your Gladia transcription request. Each entity comes back with its type, text, and exact timestamps in the audio, so you can highlight it in the transcript or jump straight to that moment in the recording.

What data can be extracted from sales call transcripts?

Gladia detects 60+ entity types, including contact names, companies, places, dates, and financial identifiers, all listed in the named entity recognition docs. A few business and medical types, such as PRODUCT and MEDICAL_CODE, are currently available for English only.

Does Gladia support real-time entity extraction?

Yes. With real-time transcription, Gladia sends a named entity recognition event for each finalized utterance, so voice agents can act on names, dates, and times mid-conversation.

Which speech-to-text platform is best for entity recognition on financial calls?

Entity extraction accuracy depends first on transcription accuracy, since a misheard digit or name can’t be labeled correctly. Solaria-3 has the lowest WER on real customer calls and financial audio out of every provider tested, and NER runs in the same request, which matters most when you’re extracting names, account numbers, and intents at scale.

Does named entity recognition work in languages other than English?

Most entity types work across languages. English has full support, and multilingual support for some types is still in progress. When callers switch languages mid-conversation, code-switching keeps the transcript accurate across the 100+ supported languages.

What’s the difference between named entity recognition and PII redaction?

Named entity recognition labels entities and leaves the transcript unchanged, while PII redaction replaces sensitive entities with masks or markers. Both share the same entity types, so you can extract data for your CRM and store a redacted version for compliance from one request.

Does named entity recognition cost extra?

No. Named entity recognition is included in Gladia’s per-hour rate for both async and real-time, with no add-on fee on any pricing plan.

Audio intelligence

Named entity recognition

Run Gladia’s named entity recognition (NER) on every transcript to extract names, companies, account numbers, dates, and medical terms, each labeled and timestamped. Turn it on in the same request as transcription for real-time streams or recorded files. 60+ entity types available.

Get your API keyGet your API key

50€

Transcription credits

Test NER on your own calls and meetings. No expiry, no credit card.

Accuracy on real customer audio

9.6% WER, the lowest of 5 providers tested, so every entity starts from an accurate transcript.

<300ms

Latency

Entities are extracted from live audio as the conversation happens, so your agents can act on them mid-call.

Trusted by over 350,000 users and 2,000+ enterprise teams

60+ entity types, ready-to-use

Skip building your own entity taxonomy. Our entity recognition covers the categories voice products need most, from contact details to clinical terms, and returns each entity with its type, text, and timestamps.

Named entity categories
Category	Example entity types	Example from a call
People and contact	`NAME` `EMAIL_ADDRESS` `PHONE_NUMBER` `USERNAME`	“You can reach Maria Lopez at maria@company.com”
Organizations and products	`ORGANIZATION` `PRODUCT` `OCCUPATION`	“We moved from Zendesk to the Enterprise plan”
Places	`LOCATION_ADDRESS` `LOCATION_CITY` `LOCATION_COUNTRY` `LOCATION_ZIP`	“Ship it to 145 Windsor Street, Toronto”
Dates and amounts	`DATE` `TIME` `DURATION` `MONEY`	“Let’s meet Friday at 3pm to review the $40,000 renewal”
Financial identifiers	`ACCOUNT_NUMBER` `CREDIT_CARD` `BANK_ACCOUNT` `NUMERICAL_PII`	“My order number is 88-2041”
Health information	`CONDITION` `DRUG` `DOSE` `MEDICAL_PROCESS`	“She takes 200 mg of ibuprofen after the CT scan”

See the full entity list in the docs

Best on real English audio

Named entity recognition is only as accurate as the transcript underneath it. On Gladia's internal English dataset of real customer calls annotated by humans, Solaria-3 hits 9.6% WER, at the top of the field and 26% better than Solaria-1.

Deepgram: 10.7% WER

"hello ladies and gentlemen thank you for standing by for cugen's third quarter twenty twenty one earnings…"

Solaria-3: 9.6% WER

"Hello, ladies and gentlemen. Thank you for standing by. Qudian's third quarter 2021 earnings conference…"

Company name mangled, numbers written as words. Those are the entities NER never gets a chance to catch.

See the full benchmark

Add named entity recognition in one parameter

Drop the second NLP service from your stack. Named entity extraction runs in the same request as transcription, so entities arrive already aligned to the audio.

Send your audio

Upload a file, pass a URL, or open a real-time WebSocket session.

Enable NER

Set named_entity_recognition to true. For live sessions, add it under realtime_processing and turn on real-time processing events.

Get labeled entities

Each entity comes back with its type, text, and start and end time, in the async result or as a live event per utterance.

Read the named entity recognition docs

Named entity recognition or PII redaction?

Pick the feature by what you need to do with the entity. Both use the same entity taxonomy, so you can detect and redact in one request.

NER vs PII redaction
Criteria	Named entity recognition	PII redaction
What it does	Labels entities and returns them as structured data	Replaces entities in the transcript with masks or markers
Transcript text	Stays unchanged	Sensitive values removed
Output	Entity type, text, and timestamps	Redacted transcript, utterances, and words
Live audio	Supported, one event per utterance	Pre-recorded audio only
Best for	CRM auto-fill, routing, search, and analytics	Compliance, safe storage, and LLM pipelines

Read about PII redaction

Use cases

What teams build with named entity recognition

Turn unstructured conversations into fields your product can act on, from CRM records to clinical notes.

Meeting assistants

Pull out people, companies, dates, and amounts to build follow-ups and CRM updates without manual notes.

Learn more →

Contact centers (CCaaS)

Capture account numbers, order IDs, and product names from every call to auto-fill tickets and route customers faster.

Learn more →

Voice agents

React to names, dates, and times as they’re spoken, so the agent can book, look up, or confirm without asking twice.

Learn more →

Extracting key data from calls?

Follow a practical approach to pulling names, account numbers, and intents from call audio with high precision.

Read the guide →

There’s a lot more than one can get out of audio than just transcription, and Gladia understood that. Feature rollouts are very proactive, and anticipate our needs as a platform. Their API performs very well with noisy telephony audio and stereo files and does an excellent job with languages.

Named entity recognition built in, no added fee

Extract entities from every recording for the same rate. Named entity recognition is included in every plan at no extra per-feature cost, for both async and real-time.

Starter

Flexible pay-as-you-go for moderate audio volumes. Get started immediately.

Async at $0.61/hr

Real-time at $0.75/hr

* 50€ in free credits

Get started

Growth

Lower unit pricing for fast-growing teams. Commit upfront to unlock savings.

Async as low as $0.20/hr

Real-time as low as $0.25/hr

* 67% less than Starter

Enterprise

Annual plan with custom models, fine-tuning, debundled pricing, and more.

Custom

Explore the pricing

Turn every conversation into structured data

Start free with 50€ in credits, or book a demo to test named entity recognition on your own audio.

FAQs

Named entity recognition is the process of finding and labeling key information in a transcript, such as names, organizations, dates, and account numbers. In voice products, NER on call transcripts combines machine learning with pattern matching to catch both free-form names and structured identifiers.
Set named_entity_recognition to true in your Gladia transcription request. Each entity comes back with its type, text, and exact timestamps in the audio, so you can highlight it in the transcript or jump straight to that moment in the recording.
Gladia detects 60+ entity types, including contact names, companies, places, dates, and financial identifiers, all listed in the named entity recognition docs. A few business and medical types, such as PRODUCT and MEDICAL_CODE, are currently available for English only.
Yes. With real-time transcription, Gladia sends a named entity recognition event for each finalized utterance, so voice agents can act on names, dates, and times mid-conversation.
Entity extraction accuracy depends first on transcription accuracy, since a misheard digit or name can’t be labeled correctly. Solaria-3 has the lowest WER on real customer calls and financial audio out of every provider tested, and NER runs in the same request, which matters most when you’re extracting names, account numbers, and intents at scale.
Most entity types work across languages. English has full support, and multilingual support for some types is still in progress. When callers switch languages mid-conversation, code-switching keeps the transcript accurate across the 100+ supported languages.
Named entity recognition labels entities and leaves the transcript unchanged, while PII redaction replaces sensitive entities with masks or markers. Both share the same entity types, so you can extract data for your CRM and store a redacted version for compliance from one request.
No. Named entity recognition is included in Gladia’s per-hour rate for both async and real-time, with no add-on fee on any pricing plan.

Named entity recognition

Transcription credits

Accuracy on real customer audio

Latency

60+ entity types, ready-to-use

Best on real English audio

Add named entity recognition in one parameter

Send your audio

Enable NER

Get labeled entities

Named entity recognition or PII redaction?

What teams build with named entity recognition

Meeting assistants

Contact centers (CCaaS)

Voice agents

Extracting key data from calls?

Named entity recognition built in, no added fee

Starter

Growth

Enterprise

Go beyond entity extraction

Custom vocabulary

PII redaction

Sentiment analysis

Turn every conversation into structured data

FAQs