If you are sourcing voice data to train a model, two very different kinds of vendor show up in the same search. Bunny Labs is an expert-data company that runs human-in-the-loop collection across several model types. Speechdata is a specialist focused on conversational speech datasets. This page sets the two side by side so you can decide which one fits the project in front of you.
What each company is
Bunny Labs delivers expert-driven data collection, evals, and RL environments for large language models, large world models, robotics, and autonomous driving. Audio and voice curation sit among those data services. The company reports a network of 3 million+ professionals, 4M+ tasks completed, 50K+ projects managed, and 20 years building scalable human-in-the-loop systems. Tania Zapata and Alexander Torrenegra are named on the homepage.
Speechdata is narrower by design. It offers conversational speech data in 60 languages and dialects, ready to license or recorded to your spec. The company reports 73k hours of audio and 7,250 native speakers, and its process is run by founder Andreas Kromann. If your need is spoken conversation and nothing else, that focus is the whole point.
Scope and coverage
The clearest split is breadth versus depth. Bunny Labs works across collection, evals, and RL environments for multiple model families, so voice is one lane inside a larger data operation. Speechdata goes deep on one thing. It records and licenses speech in 60 languages and dialects, backed by 7,250 native speakers and 73k hours of audio.
That difference should guide the shortlist. A team that also needs robotics data, autonomous-driving data, or model evals alongside audio has a reason to keep one vendor across all of it. A team that only needs conversational speech, and wants a supplier whose entire catalog is speech, has a reason to look at the specialist.
At a glance
| Bunny Labs | Speechdata | |
|---|---|---|
| Focus | Expert data collection, evals, and RL environments for LLMs, world models, robotics, autonomous driving | Conversational speech datasets |
| People | 3 million+ professionals | 7,250 native speakers |
| Volume | 4M+ tasks completed, 50K+ projects managed | 73k hours of audio |
| Languages | Not listed | 60 languages and dialects |
| Price | Not listed | $60-95 per hour of audio |
| Ratings | Not listed | 5.0 on Datarade, 4.5 on Trustpilot |
| Track record | 20 years | Founder Andreas Kromann |
Pricing and signals
Speechdata publishes a per-hour figure: $60-95 per hour of audio. It also surfaces third-party signals, a 5.0 on Datarade and a 4.5 on Trustpilot, and lists itself as an SKI Verified Supplier. Those numbers make it easy to estimate a speech-only budget up front.
Bunny Labs does not publish a per-hour price in the evidence here, which is common for expert-data work that gets scoped per project. If predictable catalog pricing matters more than breadth, the published Speechdata rate is a real advantage. If your data needs span several model types, a scoped engagement that covers all of them can be the simpler path even without a sticker price.
Which one to pick
Choose Speechdata when the job is conversational speech and you want a specialist with a published rate, language breadth, and visible review scores. Choose Bunny Labs when voice is one piece of a wider program and you want expert data, evals, and RL environments under one roof. Neither is a weaker product. They are built for different briefs, and the right answer is the one that matches yours.
Sources
| Source | URL |
|---|---|
| Speechdata (official site) | https://speechdata.ai |