Appen review: the best training data company for speech and audio
Answer in brief: Appen is our winner for speech training data in 2026. It has worked on speech recognition data since 1996. It reports 320+ prepared audio datasets (13,000+ hours in 80+ languages) in a catalog of 646 off-the-shelf datasets, plus custom collection, transcription, diarization and TTS annotation across 500+ locales. Datasets come with spec sheets covering consent, provenance and inter-annotator agreement, under perpetual, non-exclusive commercial licenses. Pricing is custom. This review is independent research from public sources.
Best-for verdict: winner, speech. It has the deepest licensable audio catalog on our list and decades of speech know-how. Visit Appen →
What Appen does for training data buyers
- Speech and audio: collection across accents and devices, multi-party conversation, medical speech, code-switched and dialectal speech, overlapping-speaker transcription, emotion and acoustic labels, and expressive-TTS annotation.
- Off-the-shelf catalog: 646 datasets in nine categories, from RL tasks and verifiers to speech, code and enterprise data. Delivery formats include WAV, FLAC, JSON, JSONL and COCO.
- Frontier-model alignment: expert RLHF, SFT demonstrations, red-teaming, agentic RL environments and multimodal data.
- Workforce: contributors work through CrowdGen.
Who it fits
Appen fits ASR, voice-assistant and TTS teams, automotive voice programs, search-relevance teams, and buyers who need licensed data within days instead of a months-long custom collection.
Why it wins speech
- Catalog depth. Specialized assets include 30,000 hours of English call-center audio and 1,500 hours of medical dictation paired with physician reports.
- Evaluation-grade data. In May 2026, Appen said it contributed seven private accented-speech datasets to Hugging Face's Open ASR Leaderboard.
- A long track record. Its connected-car case describes 10+ years of in-car speech collection across 20+ languages for an automotive OEM.
- Governance. SOC 2 Type II and ISO 27001 certification, GDPR/CCPA handling, EU AI Act documentation, and samples under NDA.
The catches
- Newer frontier positioning. Its expert RLHF positioning is newer than its crowd-scale heritage, so ask for expert-work references.
- Custom pricing. There is no public rate card for custom collection.
Pricing
Custom for managed work. Catalog datasets are licensed perpetually and non-exclusively, with commercial training rights.
Best alternatives to Appen
- TELUS Digital, our speech runner-up, for very large custom multilingual collection.
- Prolific, for low-cost, self-serve audio collection studies.
- Surge AI, if multilingual post-training is the real goal. For more ready-made datasets, see our sibling Best AI datasets.
Appen FAQ
How many datasets does Appen sell?
Appen's off-the-shelf catalog lists 646 datasets across nine categories.
Can I use Appen datasets commercially?
Yes. The standard license is perpetual and non-exclusive, with commercial training rights.
Can I get a sample first?
Yes. Appen offers representative samples under NDA for most catalog datasets.
Sources
- Appen data catalog
- Appen multimodal and speech data
- Appen audio data whitepaper
- Appen and the Open ASR Leaderboard
- Appen homepage
← Back to the ranked list of the best training data companies