What is async speech-to-text API?

Async speech-to-text is an API that transcribes pre-recorded audio or video files, returning a complete, accurate transcript once processing finishes. It's the right choice for recordings like meetings, podcasts, and support tickets, where you don't need word-by-word results in real time.

How accurate is Gladia's speech-to-text API?

Both of Gladia's models rank #1 on real-life conversational speech benchmarks, including real customer recordings and noisy, overlapping audio. Solaria-1 also delivers 3x lower diarization errors than other vendors, so speaker labels stay accurate even in multi-speaker calls.

Does Gladia's async API support multiple languages in the same audio file?

Yes. Gladia's async API supports code-switching, detecting and transcribing multiple languages within the same audio file. Solaria-1 covers 100+ languages, with automatic language detection built in, while Solaria-3 is optimized for the highest accuracy on core European languages.

Does the async API include speaker diarization?

Yes, speaker diarization is built into Gladia's async transcription API, automatically labeling who said what. Powered by an industry standard pyannote Precision 2 model, Gladia's speaker diarization is #1 in the market according to the open STT benchmark, producing 3x fewer errors than other vendors, which is critical for accurate meeting summaries, call QA, and conversation intelligence.

How does Gladia avoid hallucinations in transcripts?

Gladia's models are trained and benchmarked specifically on real, noisy conversational audio rather than clean read-speech, which reduces the model's tendency to invent words or phrases that were never spoken. This matters most for downstream uses like AI meeting summaries and compliance review, where invented details can't be tolerated.

Is Gladia's async transcription API compliant for regulated industries?

Yes, Gladia is SOC 2 Type 2, HIPAA, GDPR, and ISO 27001 compliant, with data residency options in both the US and EU. Regulated industries can keep audio and transcripts within their required jurisdiction. On-premise deployment is also available for both Solaria-3 and Solaria-1 for teams that need full infrastructure control.

How much does Gladia's async speech-to-text cost?

Gladia's async transcription starts at 0.16€/hour with Solaria-3 and 0.27€/hour with Solaria-1 on the pay-as-you-go Self-serve plan, and the rate drops automatically as your monthly volume grows, down to 0.12€/hour. New accounts get 50€ in free credits, and Enterprise offers custom pricing with fine-tuned models.

Can Gladia's transcripts be integrated with LLMs?

Yes, Gladia's transcripts are built to feed directly into LLM workflows. For instance, with Audio-to-LLM, you send raw audio in a single API call and get back both the transcript and LLM-analyzed output (summaries, extracted fields, action items) through one webhook, no separate orchestration required. Because Gladia's models prevent hallucinated text, that output is reliable enough to pass straight into your LLM workflow without manual cleanup first.

What is the difference between a transcription API and a speech-to-text API?

The terms describe the same category of tool: an API that converts audio into text. "Transcription API" is more common in developer search and documentation, "speech-to-text API" in vendor marketing. Gladia's async transcription API processes pre-recorded audio and video, batch-style, at roughly 60 seconds per hour of content, and returns structured, speaker-labeled text with word-level timestamps in one call.

Can Gladia's async API process multiple audio files at once?

Yes. Each file is submitted as a separate API call, and multiple files can run concurrently, up to 25 concurrent async jobs on the Self-serve plan, with no concurrency limit on Enterprise. Each file processes independently, at roughly 60 seconds per hour of content, up to 135 minutes and 1,000MB per file.

Async

Async transcription API

Gladia's transcription API turns pre-recorded audio and video into structured, speaker-labeled text across 100+ languages, with diarization benchmarked at up to 3x lower DER and automatic language detection. Built to hold up on noisy recordings and heavy accents.

Get €50 in transcription credits. No expiry.

On real-life speech

9.6% WER on real conversational audio, the lowest among 8 providers.

Speaker diarization

3x fewer speaker errors than leading APIs, and the lowest error rate of any provider.

100+

Languages covered

Detects and transcribes mid-sentence language switches across every supported language.

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

Async

How the async transcription API works

Gladia's async transcription API processes pre-recorded audio and video files up to 135 minutes and 1,000MB, at roughly 60 seconds per hour of content. Each file returns a structured transcript with word-level timestamps, speaker labels, and the audio intelligence features you turn on, from diarization to translation, summarization, and named entity recognition. For lower-latency needs, the Real-time API streams live audio instead.

Fewer errors than any other model

Gladia's Solaria models outperform every provider on Switchboard, the toughest conversational benchmark. Our benchmark methodology is open-source, so you can reproduce the results.

Compare models Compare models

33.9%

Solaria-3

37.3%

Solaria-1

42.3%

AssemblyAI

46%

Speechmatics

48.1%

Mistral

49.8%

Deepgram

55.2%

ElevenLabs

ACCURACY

The most accurate speech-to-text API

Our Solaria-3 leads on European business audio; Solaria-1 maximizes coverage across 100+ languages. Both benchmarked on real customer recordings.

#1 on Switchboard, the toughest conversational benchmark
Lowest word error rate among 8 providers on customer audio
3x fewer speaker diarization errors than leading APIs
Custom vocabulary for product names, acronyms, and domain-specific terms

Accurate async transcription with speaker diarization and custom vocabulary

MULTILINGUAL

Works in any language

Gladia covers languages no other provider transcribes, and handles mid-sentence language switches without any manual configuration.

100+ languages, so no call falls outside your coverage
Auto language detection, no config needed
Translation to target languages as a built-in output
Broad dialect and accent coverage

Multilingual transcription and translation across 100+ languages

COMPLIANCE

SOC 2, GDPR, and HIPAA-ready security

Sensitive audio needs more than a privacy policy. Gladia is fully compliant with requirements that matter in regulated industries.

SOC 2 Type 2, HIPAA, GDPR, ISO 27001 certified
EU and US data residency options
Your data is never used to train models on paid plans

Enterprise security certifications including GDPR, SOC2, and HIPAA

Everything you need, already built in

Skip the tool-stitching. Get raw audio to accurate data in one place.

Speaker diarization

Automatically detect and label who said what in multi-speaker audio, with 3x fewer errors than other vendors.

Learn more

Custom vocabulary

Boost recognition accuracy for product names, acronyms, and domain-specific terms with keyterm prompting.

Learn more

Named entity recognition

Pull out the names, dates, and organizations that matter, turning raw speech into structured, queryable data.

Learn more

Speech-to-text translation

Translate audio into 100+ languages, keeping speaker labels and timestamps intact.

Learn more

Audio-to-LLM

Turn audio into structured insights, including summaries and action items, in a single API call, with access to 400+ LLM models.

Learn more

PII redaction

Automatically mask names, emails, phone numbers, and other sensitive data before it ever leaves your pipeline.

Learn more

See the full list See the full list

Match the model for your workflow

Solaria-3 and Solaria-1 are built for different jobs:
Solaria-3 for the highest accuracy on European and English audio. Solaria-1 for the broadest coverage.

Solaria-3

Highest accuracy on European real-world audio

#1 on European business audio and conversational speech
Optimized for EN, FR, DE, ES, IT
#1 on English production audio

Learn more

Solaria-1

Maximum language coverage across any domain

The most multilingual model: 100+ languages
3x lower diarization errors than other vendors
Code-switching, handled natively

Learn more

We are 100% benchmark and evaluation driven. Gladia was one of the best providers selected on merit to transcribe user videos, especially for non-English languages. Their reactive customer support and data compliance make their offer really compelling.

USE CASES

Built for real world use cases

Teams use Gladia's async transcription API to turn raw audio into accurate, structured, and speaker-labeled data they can trust.

Meeting assistants

Accurate, speaker-labeled transcription that powers reliable AI summaries, action items, and CRM sync, without the model inventing details that were never said.

Learn more

Contact centers (CCaaS)

Transcribe and analyze recorded calls at scale for QA, compliance review, and agent coaching, with accuracy that holds up on compressed, noisy 8kHz phone audio.

Learn more

Media & content

Generate accurate, time-stamped subtitles and transcripts for video and podcast content, with support for multi-channel audio.

Learn more

Something else entirely

If your product records, transcribes, or understands audio in any context, Gladia handles the layer underneath. Explore the API and build what you have in mind.

View our docs

Compare pricing

Pay only for the audio you transcribe. Async rates drop automatically as your monthly volume grows.

Self-serve

Flexible pay-as-you-go for moderate audio volumes

Starts at

0.16€/hour

Start with 50€ free credits

Enterprise

Annual plan with custom models, fine-tuning, and more

Custom

Top-up & save

Earn free credits when you top up

Free credits by top-up amount
Top up	Bonus	You get
1,000€	+10%	100€ in free credits
2,500€	+15%	375€ in free credits
5,000€	+20%	1,000€ in free credits

See our pricing

Start transcribing in minutes

FAQs

Async speech-to-text is an API that transcribes pre-recorded audio or video files, returning a complete, accurate transcript once processing finishes. It's the right choice for recordings like meetings, podcasts, and support tickets, where you don't need word-by-word results in real time.
Both of Gladia's models rank #1 on real-life conversational speech benchmarks, including real customer recordings and noisy, overlapping audio. Solaria-1 also delivers 3x lower diarization errors than other vendors, so speaker labels stay accurate even in multi-speaker calls.
Yes. Gladia's async API supports code-switching, detecting and transcribing multiple languages within the same audio file. Solaria-1 covers 100+ languages, with automatic language detection built in, while Solaria-3 is optimized for the highest accuracy on core European languages.
Yes, speaker diarization is built into Gladia's async transcription API, automatically labeling who said what. Powered by an industry standard pyannote Precision 2 model, Gladia's speaker diarization is #1 in the market according to the open STT benchmark, producing 3x fewer errors than other vendors, which is critical for accurate meeting summaries, call QA, and conversation intelligence.
Gladia's models are trained and benchmarked specifically on real, noisy conversational audio rather than clean read-speech, which reduces the model's tendency to invent words or phrases that were never spoken. This matters most for downstream uses like AI meeting summaries and compliance review, where invented details can't be tolerated.
Yes, Gladia is SOC 2 Type 2, HIPAA, GDPR, and ISO 27001 compliant, with data residency options in both the US and EU. Regulated industries can keep audio and transcripts within their required jurisdiction. On-premise deployment is also available for both Solaria-3 and Solaria-1 for teams that need full infrastructure control.
Gladia's async transcription starts at 0.16€/hour with Solaria-3 and 0.27€/hour with Solaria-1 on the pay-as-you-go Self-serve plan, and the rate drops automatically as your monthly volume grows, down to 0.12€/hour. New accounts get 50€ in free credits, and Enterprise offers custom pricing with fine-tuned models.
Yes, Gladia's transcripts are built to feed directly into LLM workflows. For instance, with Audio-to-LLM, you send raw audio in a single API call and get back both the transcript and LLM-analyzed output (summaries, extracted fields, action items) through one webhook, no separate orchestration required. Because Gladia's models prevent hallucinated text, that output is reliable enough to pass straight into your LLM workflow without manual cleanup first.
The terms describe the same category of tool: an API that converts audio into text. "Transcription API" is more common in developer search and documentation, "speech-to-text API" in vendor marketing. Gladia's async transcription API processes pre-recorded audio and video, batch-style, at roughly 60 seconds per hour of content, and returns structured, speaker-labeled text with word-level timestamps in one call.
Yes. Each file is submitted as a separate API call, and multiple files can run concurrently, up to 25 concurrent async jobs on the Self-serve plan, with no concurrency limit on Enterprise. Each file processes independently, at roughly 60 seconds per hour of content, up to 135 minutes and 1,000MB per file.