Appen is one of the best-known names in AI training data. Its catalog lists speech recordings that span read, conversational and contact-center audio, and it also collects and annotates speech to order. Many teams buy from Appen, and many compare it with other vendors before they sign. This guide is for the second group.
It covers eight alternatives for licensed speech data, what each offers for ASR, TTS and voice AI training, and who it tends to suit. A note on bias: we run SpeechData.ai, which is one of the eight. Everything about the other vendors comes from their own websites and press releases, checked in October 2026. We left out prices unless a vendor publishes them.
Why buyers compare Appen alternatives
The reasons are usually about fit rather than about any one vendor.
- Published prices. Some teams want a price per hour of audio before the first sales call, to size a budget or get sign-off.
- A specific language or dialect. Catalogs differ, and the dialect you need may be deep at one vendor and thin at another.
- Conversational or scripted speech. Voice agents and call analytics need spontaneous dialogue. Wake words and TTS need controlled recordings. Vendors lean one way or the other.
- Order size. A few hundred hours of one language and a multi-year program across dozens of locales are different purchases, and vendors are set up differently for each.
- Consent and license paperwork. Legal teams want the consent wording and the license scope in writing before data enters a training run.
- A second source. Some teams keep a second vendor to compare quality and avoid depending on one supplier.
Appen alternatives at a glance
| Vendor | Speech data offer | Catalog or custom | Prices on its site |
|---|---|---|---|
| DataForce (TransPerfect) | Scripted and conversational speech, wake words, in-car audio and studio TTS, collected to order; transcription | Custom | No |
| Defined.ai | Marketplace of speech datasets, including large call-center sets; custom collection and transcription | Both | No |
| FutureBeeAI | Conversation, call-center, scripted, wake-word and TTS speech sets by language and domain; custom collection | Both | No |
| LXT | Conversational, command, dictation and read speech collected to order; transcription | Custom | Indicative range for custom projects |
| Nexdata | Dialogue and read speech recognition datasets and TTS datasets; custom collection and annotation | Both | No |
| Shaip | Call-center, conversation, command, scripted, dictation and TTS datasets; custom collection | Both | No |
| SpeechData.ai | 60 conversational datasets totaling 73,000 hours, dual-channel, with consent documentation; custom collection | Both | Yes, per hour of audio |
| TELUS Digital | Managed annotation and collection programs; a small catalog with a few audio datasets | Mostly custom | No |
The alternatives, one by one
DataForce (TransPerfect)
DataForce is the AI data division of TransPerfect, the language services company. We found no off-the-shelf speech catalog on its site; its speech work is collected to order. It describes scripted or conversational speech, wake words, multi-speaker dialogues and in-car recordings, records TTS data in TransPerfect's own studios, and offers speech and audio transcription. It cites more than 250 languages and dialects and over a million contributors.
Suits: teams that need speech recorded to a detailed spec, such as particular devices, in-car conditions or a studio voice, from one vendor that also transcribes it.
Defined.ai
Defined.ai, formerly DefinedCrowd, runs a marketplace where companies buy and sell AI training data. Its speech pages cite more than 2 million hours of speech data in over 500 languages and dialects, from scripted monologues to live call-center and podcast audio, and it also collects speech to order and offers transcription and diarization. Some large telephony sets list transcription as a separate service, so check what each listing includes. Dataset pages offer samples and quotes on request.
Suits: buyers who want to browse a large marketplace, including very large call-center corpora, and are comfortable requesting a quote per dataset.
FutureBeeAI
FutureBeeAI sells ready-made speech datasets organized by language and domain: general and call-center conversations, scripted monologues, wake words and commands (including in-car) and TTS. Doctor dictation and doctor-patient conversation sets are listed as coming soon. Sets are sized per language and domain; one Egyptian Arabic financial services call-center set, for example, lists 40 hours from 80 participants. It also runs custom collection, transcription and audio annotation.
Suits: teams that want a focused set for one language and one domain, such as call-center audio for a single industry.
LXT
LXT focuses on speech collected to order. Its speech pages describe spontaneous conversation, commands, dictation and read speech, telephone recordings, in-car and wake-word data in more than 100 languages, plus transcription, ASR post-editing and diarization. Its homepage mentions pre-built datasets, but we found no list of ready-made speech datasets on the site. Besides us, LXT is the only vendor here with a published price guide: its speech recognition page says most custom projects range from $15K for a focused domain dataset of 50 to 200 hours to $150K+ for large multi-accent, multi-environment collections of more than 1,000 hours.
Suits: enterprises that want managed custom collection and transcription across many locales.
Nexdata
Nexdata runs one of the larger speech catalogs: in October 2026 its site listed 269 speech recognition datasets, split between dialogue and read speech, and 40 TTS datasets, with prices quoted on request. It also collects conversational and children's speech to order and offers speech annotation.
Suits: buyers who want to license existing ASR or TTS corpora in many languages quickly, including read speech.
Shaip
Shaip licenses off-the-shelf speech datasets and collects to order. Its catalog groups sets into call-center, general conversation, voice command, scripted monologue, dictation and physician, and TTS, labeled by sample rate and shipped with transcriptions. Prices are quoted per audio hour on request. In February 2026, Ubiquity announced an agreement to acquire Shaip.
Suits: buyers who want existing call-center, command or medical speech in a specific language.
SpeechData.ai
SpeechData.ai specializes in conversational speech and publishes its prices. The catalog has 60 datasets, one per language or dialect, with 500 to 2,000 hours each of spontaneous two-person conversation recorded in-country. Every dataset is dual-channel, with time-aligned, human-verified transcripts, age, gender, region and device metadata for every speaker, and the consent each speaker signed. Prices run from $60 to $95 per hour of audio, and the pricing page breaks them down by region with worked examples.
Free samples go out on request to a company email. Languages, domains, devices or environments outside the catalog can be collected to spec, scripted or spontaneous.
Suits: teams that need natural dialogue for ASR, voice agents or speech LLMs and want to know the price before the first call.
Where a larger vendor may fit better: the catalog has no read speech, wake-word or TTS studio datasets off the shelf, it covers 60 languages and dialects rather than hundreds, and we are a small specialist team. If you want speech alongside text, image and other data work under one enterprise contract, a full-service vendor is the simpler choice.
TELUS Digital
TELUS Digital, formerly TELUS International, bought Lionbridge's AI division at the end of 2020 and has been wholly owned by the Canadian telecom company TELUS since October 2025. Its AI data business runs large annotation and collection programs: its site lists speech transcription, speaker identification, diarization and audio segmentation, and one case study describes more than 4 million audio prompts and utterances from over 5,000 contributors for a voice assistant. A small off-the-shelf catalog includes a few speech datasets, such as US English, UK English and German.
Suits: large enterprises that want a managed data program in which speech is one part.
How to compare your shortlist
Send every vendor the same short spec: languages and dialects, hours, conversational or scripted speech, channels, required metadata, transcription conventions and how the model will be used. Then compare four things side by side.
- A price per hour of audio in writing, and what it includes: transcripts, metadata, license term and delivery format.
- A free sample you choose, listened to and run through your current model.
- The consent wording speakers signed, and a license that covers training and deploying a commercial model.
- Transcription conventions and folder layout, so new data lines up with what you already have, including anything bought from Appen.
Our comparison of speech data providers has a longer checklist and also covers free open datasets.
