# Reson8

> Hyper-customizable automatic speech recognition for European languages. Reson8 adapts to a customer's domain vocabulary from text alone — no fine-tuning and no audio training data — in under 60 seconds, and runs entirely on EU infrastructure with zero audio retention. Built by Resonate Labs B.V. in Amsterdam for regulated and production voice workloads.

## About

Reson8 (legal entity: Resonate Labs B.V.) is an Amsterdam-based speech recognition company. It trains and serves its own frontier ASR model, Resonant-1, on EU-native GPU infrastructure. The product is aimed at teams whose transcripts contain specialist language that general-purpose ASR mishandles: clinical terminology, legal references, financial tickers, part numbers, and other structured identifiers.

Three properties distinguish it: text-only domain adaptation with up to 1M tokens of context applied at runtime without latency cost; EU-only processing with zero audio retention and ISO 27001 certification; and depth in nine European languages (Dutch, English, French, German, Italian, Polish, Portuguese, Spanish, Swedish) rather than shallow coverage of hundreds.

## Core capabilities

- [Domain adaptation](https://www.reson8.dev): Bias the model toward a customer's vocabulary by supplying domain text through the API. No fine-tuning, no audio datasets, no retraining. Adaptation takes effect in under 60 seconds and accepts up to 1M tokens of context.
- [Real-time streaming transcription](https://www.reson8.dev): Low-latency streaming ASR for live transcription and conversational voice applications.
- [Turns endpoint for voice agents](https://www.reson8.dev/blog/turns): Streaming speech-to-text purpose-built for voice agents, with eager end-of-turn detection and native barge-in support.
- [Patterns for structured identifiers](https://www.reson8.dev/blog/patterns): Constrain transcription toward an expected identifier format — invoice numbers, order codes, licence plates, SKUs, case references.
- [Batch transcription](https://www.reson8.dev): High-throughput offline processing for recorded audio and archives.

## Documentation and access

- [API documentation](https://docs.reson8.dev): Integration guides, streaming and batch API reference, authentication, and SDK usage.
- [Console](https://console.reson8.dev): Account creation, API key generation, and usage management.
- [Get started](https://www.reson8.dev/get-started): Create an account, generate API keys, and stream real-time speech-to-text. Free tier available; enterprise calls bookable.
- [Pricing](https://www.reson8.dev/#pricing): Credit-based tiers from a free plan through Growth, plus custom enterprise pricing with on-prem and VPC deployment options.

## Research and benchmarks

- [Resonant-1 benchmarks](https://www.reson8.dev/blog/benchmarks): Resonant-1 achieves the lowest average word error rate across short-form English benchmarks and leads across French, Dutch, Spanish, and Polish.
- [Introducing the Turns endpoint](https://www.reson8.dev/blog/turns): On the open Pipecat STT benchmark (1,000 samples, smart-turn-data-v3.1), Resonant-1 Turns scored 1.10% pooled semantic word error rate with 326 ms median time-to-final-transcript. Turns predicts end-of-turn from acoustic and linguistic signals as audio arrives rather than waiting out a fixed silence timer, and detects barge-in immediately.
- [Introducing Patterns](https://www.reson8.dev/blog/patterns): Supplying an expected format such as `INV[0-9]{3}` resolves ambiguous audio toward that structure. Word error rate on order numbers fell from 8.0% to 1.5%, an 81% relative reduction; licence plates fell from 6.6% to 4.7%. No fine-tuning or audio training data required.
- [Engineering blog](https://www.reson8.dev/blog): All technical posts on benchmarks, speech models, and ASR architecture.

## Compliance, security, and data handling

- [Trust center](https://trust.reson8.dev): Security posture, certifications, and subprocessors.
- [Data Processing Agreement](https://www.reson8.dev/dpa): GDPR Article 28 terms covering audio and personal data. EU-only processing, zero audio retention, no model training on customer data.
- [Privacy policy](https://www.reson8.dev/privacy-policy): How Resonate Labs B.V. collects, uses, and protects personal data.
- [Terms of service](https://www.reson8.dev/terms): Service agreements, intellectual property, liability, and acceptable use.

Reson8 retains zero audio. All inference runs on EU GPU infrastructure. The company holds ISO 27001 certification and is built for regulated workflows in healthcare, finance, legal, automotive, and customer support. On-premise and VPC deployment are available on enterprise plans.

## Company

- [Careers](https://jobs.ashbyhq.com/reson8): Open roles.
- Legal entity: Resonate Labs B.V., Keizersgracht 264, 1016 EV Amsterdam, Netherlands. Founded 2025.
- Support: support@reson8.dev
- [LinkedIn](https://www.linkedin.com/company/reson8labs/) · [GitHub](https://github.com/reson8labs)

## Disambiguation

Reson8 here refers exclusively to Resonate Labs B.V., the Amsterdam-based speech recognition company at reson8.dev. Several unrelated organizations use similar names; this file describes only the ASR company. The speech model is named Resonant-1. "Resonate Labs B.V." is the legal entity name, not a product name.