Set per task, shown up front
Before you accept any task, you see exactly what it pays. The price is agreed up front and does not change after you submit.
Spirelight recruits people for paid remote AI training work in four shapes: voice recording, transcription, transcript verification, and data annotation. No experience, degree or coding background is required. Joining is free, every task carries a fixed price you see before you accept it, and projects open in waves, so signing up does not guarantee immediate work.
These are remote AI training jobs in the literal sense: you produce or check the data a speech model learns from. Nothing here needs machine learning knowledge. It needs a quiet room, a clear voice or a careful ear, and the discipline to follow a written brief exactly.
| Task type | What a real task looks like | Typical length |
|---|---|---|
| Voice recording | You read prompts from a script, or speak spontaneously on a given topic, into your phone or laptop. The brief sets the room conditions, the distance from the microphone, and how many takes are allowed. Rejected clips usually fail on background noise, not on your delivery. | 5 to 30 minutes per batch |
| Transcription | You type out what is said in a recording, following a style guide that tells you how to handle hesitations, overlapping speech, numbers, and words in another language. Accuracy against the guide matters more than speed. | 4 to 8 minutes per audio minute |
| Transcript verification | You listen against an existing transcript, often one produced by a model, and correct what is wrong. Slower and more attentive than it sounds, because the errors are the parts a machine found hard. | 2 to 4 minutes per audio minute |
| Data annotation | You apply labels to audio or text against a taxonomy: who is speaking, where a speaker changes, what the intent is, whether a clip is usable. Data annotation jobs are where consistency across hundreds of small decisions earns the repeat invitations. | 10 to 40 seconds per item |
Times are typical ranges from current projects, not commitments. Every task brief states its own scope and its own price. If you are new to the field, read what an AI trainer does before you sign up.
Tasks are priced individually and shown before you accept. Across current projects that works out to roughly 12 to 50 USD per hour, depending on the task type and how scarce your language is.
The bottom of that band is routine work: short prompts read from a script, straightforward transcription in a widely spoken language, simple labelling passes. The top of the band is where scarcity pays. Rare languages, strong regional accents, code-switching, specialist domains, and tasks that need a specific recording setup all move the rate up, because the pool of people who can do them is small.
Every task carries a fixed price you see before you accept it, and that price does not change after you submit. A human reviews your work, and the payout is released to your contributor account on approval. There are no invoices to chase and no unpaid trials.
Treat any platform that advertises a guaranteed rate per hour with caution, because per-task pricing is what real work in this field looks like and nobody can know your number before you have seen a task. Joining is free wherever the work is real.
Before you accept any task, you see exactly what it pays. The price is agreed up front and does not change after you submit.
Once a human reviewer approves your submission, the payout is added to your contributor account. No invoices to chase.
Creating a profile costs nothing, and there is no kit to buy. Qualification tasks, where a project uses them, are paid like any other task.
These are work from home jobs with no experience requirement, because the thing being tested is care, not credentials.
You do not need a background in AI, linguistics or software. What projects screen for is whether you read the brief, keep your recording conditions consistent, and hand back work that does not have to be fixed. People who do that get invited to the next wave.
A quiet room matters more than expensive gear. A recent phone recorded in a small room with soft furnishings will pass where a good microphone in a room with hard walls and traffic outside will not. Fixing your environment costs nothing; replacing rejected recordings costs you your time.
Demand shifts with every active project. What speech datasets are consistently short of is not more standard English: it is regional accents, dialects, code-switching between two languages in one sentence, and languages that barely appear in existing training data. List every language and dialect you speak when you build your profile, including the ones you assume nobody wants.
Language and country pages list what tends to open where, and what a profile needs to be considered.
This field has a real scam problem, so the honest answer is a set of checks rather than a reassurance. The four rows below are what Spirelight commits to. Apply the same four questions to any platform that contacts you, and walk away from the ones that will not answer them in writing.
| Check | What Spirelight commits to |
|---|---|
| Cost to join | Free, permanently. Spirelight never charges a joining fee, never sells you a kit or training course, and never asks for banking passwords, card details, or a payment to release your earnings. Any platform that does is a scam. |
| Pay visibility | The fixed price for a task is shown in the brief before you accept it, alongside the rules your work will be reviewed against. You decline anything that is not worth your time, with no penalty. |
| Review and payout | A person, not a black box, reviews what you submit. On approval the payout is released to your contributor account. If something needs a fix, you are told why and can resubmit. |
| Availability | Projects open in waves and require particular languages, locations, and equipment. Creating a profile does not guarantee immediate work. It puts you in the pool that gets contacted when a matching project opens, which may be the same week or several weeks later. |
That last row is the one most platforms leave out, and it is the one that decides whether this is worth your time. Treat AI training work as paid work that arrives in waves, not as a salary. For a wider comparison, see AI training sites that pay and how to identify legitimate AI training jobs.
Sign up once. Get matched when a project fits. Record, get reviewed, get paid.
A few minutes. Your native languages, regional dialects, accents, location and equipment are what projects are matched against, so list all of them rather than the ones you think sound impressive.
When a project matches your profile you receive the task details, the guidelines, the fixed price, and the review process up front. Accept the work or skip it freely.
Complete tasks in your browser or upload your files. Once your work clears human review, the price for each approved task is released to your account.
Payment follows approval rather than submission, so expect a gap between finishing a task and being paid for it. Review commonly takes days, and platforms then settle on a stated cycle with a minimum payout threshold. That lag is normal in this industry and is not a warning sign by itself. A platform that will not put its review and payment terms in writing before you work is the warning sign. On Spirelight each project states its terms before you accept a task.
The established platforms do pay, and most complaints you will read are about task volume and slow support rather than about work going unpaid. Protect yourself with process rather than reputation: keep your own record of what you submitted and when, check the stated approval window before you accept work, and chase anything that passes review but stays unpaid beyond the stated cycle, in writing, straight away.
Per task, on Spirelight and on most platforms in this field. You see a fixed payout for an item before you accept it, and you are paid for work that passes review. Any hourly figure you see quoted anywhere, including ours, is an effective rate observed across completed per-task work rather than a wage anyone can promise you. That is also why accuracy pays better than speed: a submission that fails review costs the time twice and pays once.
Independent contractor work, here and on essentially every platform in this field. There is no guaranteed volume, no notice period and no holiday pay, and you are responsible for reporting your own income where you live. In exchange you choose which tasks to accept and when to work. Anyone describing this as a salaried position is describing something that does not exist in this market.
Legitimate work means you apply, complete a stated task, and get paid under the project terms. It is a scam if you are asked to pay a fee, buy a kit, or share banking passwords before you work. Spirelight never charges you to join, and submitted work is reviewed against the project rules. We wrote a full guide on how to tell real AI training jobs from the scams.
Tasks are priced individually and shown before you accept. Across current projects that works out to roughly 12 to 50 USD per hour, depending on the task type and how scarce your language is. Routine work sits at the bottom of that band and scarce languages, strong regional accents, code-switching, and specialist recording setups sit at the top.
Yes. Companies pay people to record speech, transcribe audio, verify transcripts, and label the data models learn from, and much of this work can be completed remotely. On Spirelight, the pay and terms are shown for each available project before you accept it, and completed work is reviewed before payment.
No. Most speech and data tasks need no coding and no formal qualifications. You mainly need a quiet space, a recent phone or a laptop with a decent microphone, a stable connection, and the patience to follow a brief exactly. Care and consistency matter more than a CV.
For recording work, a recent smartphone or a laptop with a clear built-in or USB microphone is often enough. A quiet room with little echo makes a bigger difference than expensive gear. Transcription and data annotation are usually easier with a computer and headphones. Each project states its own device requirements.
Demand changes with each active project. Speech datasets often need regional accents, dialects, code-switching, and languages that are underrepresented in existing data, so it is worth listing every language and dialect you speak when you build your profile. Start with the language and country guides.
The work belongs to the same broad AI data category, but Spirelight is speech-first. Projects can include recording, transcribing, checking, and labelling voice data, with matching based on the language, dialect, accent, location, and equipment a project needs. Project availability and terms vary, so compare the task brief before accepting work on any platform.
You can register and build your profile in a few minutes. Work is not guaranteed immediately because projects open in waves and require particular languages, locations, or equipment. When a matching project is available, its pay, review process, and payment terms are shown before you accept it.
Still unsure? Write to contributors@spirelight.net before you sign up.
Register once and build your speaker profile. It is free, and it takes a few minutes. Projects open in waves, so joining does not guarantee immediate work; it makes your profile available when a paid project matches your language, accent and location.