Quality assurance

Speech-to-text built for contact center quality assurance

For contact center quality assurance, a misheard disclosure fails a compliant agent and a misattributed line blames the wrong one. Gladia gets both right on real call audio.

Used by

  • Aircall
  • Selectra
  • Gravite
Contact center quality assurance illustration
93%
Less review time per call
From about 15 minutes of manual review to 1 minute, in production.
#1
On recorded phone conversations
9.6% WER on real conversational audio, the lowest among 8 providers.
#1
Speaker diarization on DIHARD III
Agent and customer separated correctly, so talk ratios and scores land on the right person.
Why teams switch to Gladia

Accurate enough to score every call. Fast enough to review same-day.

Every QA score, compliance flag, and coaching note inherits the accuracy of the transcript underneath it. It's the same speech-to-text for contact centers behind Aircall and Selectra.

Keyword accuracy

The words that decide pass or fail

QA scorecards don't fail on average WER. They fail on a missed disclosure or a wrong account number. Custom vocabulary and entity extraction get those words right.

  • Custom vocabulary for scripts & products
  • Named entity recognition
  • #1 on entity-dense Earnings22 audio
Speaker attribution

Score the right agent

Talk ratios, per-agent sentiment, and compliance attribution all depend on who said what. Gladia labels every word by speaker, on real call audio.

  • Word-level speaker labels
  • Per-speaker sentiment
  • #1 on DIHARD III
Full coverage

Score every call, not a sample

Manual QA reaches as few as 1–2% of calls. Async transcription fast enough to process every recording makes 100% coverage practical, reviewed the same day.

  • 100% call coverage
  • Fast async processing at volume
  • Same-day review
Automated scoring

Your scorecard, in one request

Transcription, speaker labels, and your own LLM scoring prompts run in a single request. Sentiment, summaries, and scorecard fields come back structured.

  • Your scorecard as a prompt schema
  • Per-speaker sentiment
  • Call summaries
Compliance & privacy

Score calls without exposing customers

Redact card numbers and customer details before transcripts reach your dashboard, with EU data residency and certifications ready for procurement.

  • 40+ redactable PII entity types
  • EU & US data residency
  • SOC 2 Type 2, ISO 27001 & 27701
Benchmarks

Tested on real phone calls, not studio audio

Switchboard is the toughest public benchmark for telephone speech: real 8kHz phone calls, crosstalk and all. Solaria-3 is the only model under 35% WER. Lower is better. Our benchmark methodology is open-source.

See the full comparisonSee the full comparison
33.9%
Solaria-3
37.3%
Solaria-1
42.3%
AssemblyAI
46%
Speechmatics
48.1%
Mistral
49.8%
Deepgram
55.2%
ElevenLabs
Getting started

From API key to scored calls in minutes

Start for free

with €50 on us

Get your API keyGet your API key
01

Get your API key

€50 in free credits, with diarization, custom vocabulary, and named entity recognition included.

02

Send your call recordings

Upload recordings for async processing from wherever they live: Genesys, Talkdesk, Twilio Flex, or any SIP/VoIP stack. Works via API, SDKs, or the Playground.

03

Receive QA-ready output

Speaker-labeled transcripts with sentiment, entities, and summaries, plus your scorecard fields through LLM prompts. Structured for your QA pipeline.

04

Benchmark on your own calls

Compare Gladia against your current provider on your real call audio, including accuracy on the entities your scorecards depend on.

Contact center QA case study

How Gravite cut call quality review time by 93%

Challenge

One misheard word could fail a call that should pass

Gravite's QA platform scores calls on exact transcript text, so a misheard greeting or missed brand name triggers a false failure. They needed accuracy they could build a product on, at volume.

93%
Less review time per call

From about 15 minutes of manual review to one.

50k
Hours of audio transcribed

With no fallback or secondary provider.

100%
Call coverage

Economically viable for the first time, instead of spot-checks.

What builders say

Trusted in production by enterprise teams

Attention
G2

Gladia's API performs very well with noisy telephony audio and stereo files.

Anis
Anis B. CEO at Attention
Aircall

The speed and accuracy improvements were game-changers. We cut transcription time by 95% and the multilingual support is unmatched.

Farid Issabhaï
Farid Issabhaï Staff Engineer at Aircall
Amanda Zhu
Amanda Zhu Co-Founder at Recall

Gladia's real-time code-switching has been a real 'wow' factor! Plus, the accuracy of transcription has been excellent.

Recall

Gladia's real-time code-switching has been a real 'wow' factor! Plus, the accuracy of transcription has been excellent.

Amanda Zhu
Amanda Zhu Co-Founder at Recall
Selectra

Gladia's feature rollouts and performance with noisy telephony and stereo audio have genuinely impressed our team.

Alexandre
Alexandre Bouju CTO Deputy Manager at Selectra

Your QA scores are only as good as the transcript they're built on

Join 2,000+ enterprise teams building their products with reliable audio intelligence.