AUDIO EVIDENCE
Audio test library
What the tests say so far
Best for: creators who want to hear voices before spending — Test 01’s identical-script samples let you compare Brian and Sarah on the same hard script, with documented settings and dates.
Skip the verdicts for now: only Test 01 is published; Tests 02–04 are protocols waiting on generation. We won’t tell you which voice “wins” until we’ve actually listened and written it down.
Bottom line: listen to Test 01 and decide with your own ears. That’s the whole point of this page.
Test 01 is published below; the rest are planned tests with their protocols already written. When each result lands, its entry carries the exact conditions: model, voice, settings, script, generation date, and what we heard — good and bad. We never publish a sample we didn’t actually generate, and we never cherry-pick an unusually bad baseline to make a fix look heroic.
Download the listening scorecard — score any two voices yourself with our one-page method (plain text, free).
What each test documents
- The specific problem being tested.
- The script and generation conditions, kept comparable across before/after.
- Model, voice, settings (stability, similarity, style), and date.
- What changed — and where the fix audibly failed or didn’t.
- Which current ElevenLabs plan supports the workflow (prices re-verified live before publishing).
Published tests
Test 01 — Same script, two voices: Brian vs. Sarah
The canonical 854-character A/B script — “$1,200”, NASA, FAQ, API, Dr. Alvarez, 2026 — rendered in both voices with identical settings. Listen for how each handles the acronyms, the dollar amount, and the pacing between sentences.
Voice A — Brian (Deep, Resonant and Comforting)
Voice B — Sarah (Mature, Reassuring, Confident)
Test 02 — Revision workflow: patching one line (planned)
Clone regeneration of a single changed line vs. the original read. Listening for audible splice points. Audio pending test completion.
Test 03 — Robotic delivery: script rewrite vs. raw text (planned)
The same voice reading a stiff script vs. a conversational rewrite. Audio pending test completion.
Test 05 — Blind listening test: 5 voices, 1 is human (in production)
Our signature series: five voices read the same script — three ElevenLabs voices we generate, one official Murf sample, one real human. Real listeners guess blind; scores publish win-or-lose. The full blind-test protocol (sample texts, scoring criteria, clip-source labeling) is public; Episode 1 results land here when the test is actually run.
Test 04 — ElevenLabs vs Murf: identical-script A/B comparison (planned)
One 150-word narration script — with the acronyms, dollar amounts, and numbers that break AI voices — rendered in both tools under documented settings. The protocol is written on the comparison page; samples embed here when generated.
Demo audio is generated on our own paid ElevenLabs account (Starter plan, commercial license active since September 2026). First recordings are produced once API access is connected — the test protocol is already written, and no sample is published until it’s actually generated.