elevenlabs.providers.sgit.ai / experiments / Voice explorer

Voice explorer

The vendor's library is large and the useful part of it is small. This lab lists what your key can reach, filters it down, and makes each candidate say your line — because a voice that sounds excellent reading a marketing sentence can still mangle a product name.

Which pattern this is — read before you paste anything

Your own full key, in your own browser, bounded only by your plan's monthly quota. That is pattern 0 with a ceiling: the narrow case where it is defensible — the key's owner, testing their own key, on their own device — and not a pattern to publish. No key ships in this page: the field below is empty until you fill it.

ElevenLabs offers no per-key spend limit for text to speech, so nothing here caps what a leaked key could cost you except the account's own quota. Use a key scoped in the dashboard to the endpoints this lab needs, and forget it when you are done. The version of this page with no key box at all is pattern three, and it does not exist yet.

1 · Load the library

Nothing loaded. GET /v1/voices is a read: it costs nothing and generates no audio.

2 · Filter

Search name, description or use case

3 · The library

	Voice	Category	Labels	voice_id	Hear it

The vendor returns a preview_url for most voices, hosted on their own CDN. This page deliberately does not fetch it: the claim that these labs call exactly one host would stop being true. Copy the link and open it yourself, or spend a cent hearing the voice say a line of your text — which is the more useful test anyway.

4 · Hear one, on your own line

The line to speak — keep it short; you pay per character

0 chars · $0.000 per voice tried

5 · Log

What this tests

Whether a voice is right for a workload is not answerable from a name and a preview clip. This lab answers three narrower questions that are:

What does this key actually reach? premade library voices, your own clones, anything shared with the account — the list is per-account, so it is a fact about your key, not about the vendor.
How does a candidate handle the words you actually use? The line in section 4 defaults to a real sentence from a real script. Swap it for the worst one you have: the one with a product name, an initialism and a version number in it.
Which three are worth a real comparison? The shortlist is kept in this browser's localStorage and exports as markdown, so the decision can be recorded somewhere durable instead of in a browser tab.

Method, and its one deliberate omission

The vendor returns a preview_url for most voices, on their own CDN. This lab does not fetch it. Every page on this site claims to call exactly one host verified 7 Sep 2026, and quietly loading audio from a second one would make that claim false. The link is there to copy; the honest test is a generation of your own text, which costs about a cent and tells you more.

Shortlisting writes nothing to the network. Listing voices is a read and costs nothing vendor docs 5 Sep 2026.

What we know, and what we do not

We know the endpoint answers from a browser, because we watched it do so verified 5 Sep 2026. We know nothing at all about which voice suits this workload: the evaluation that produced this site used the premade "Adam" voice for every sample because it was the first one in the list, which is not a choice, it is an absence of one verified 5 Sep 2026.

For narration of technical explainers, the labels worth filtering on are use_case: narration or informative_educational, and a neutral accent. That is a starting hypothesis from the vendor's own metadata written, not run, not a finding.

What it costs

Listing is free. Each "speak my line" is one generation of your text — around $0.006 for a 60-character line on flash, twice that on v3 vendor docs 5 Sep 2026. Trying eight voices on one line costs less than a bus fare and settles an argument that otherwise runs for a week.

Every claim on this site, with its state → Disclosures →