How VoxFree Evaluates Voices and Product Claims
This page separates machine checks, human judgment, product facts, and competitor facts so a reader can see what each claim is based on.
Voice Candidate Evaluation
- Freeze the provider, model version, locale, candidate IDs, and seven test categories before listening.
- Generate short text, numbers, names, abbreviations, a long sentence, mixed-language text, and a long-form sample for every candidate.
- Run automatic checks for successful synthesis, sample rate, channel count, duration, finite samples, clipping, and silence.
- Have a human reviewer listen and score naturalness, pronunciation accuracy, long-sentence stability, long-form consistency, audio integrity, code-switching, and use-case fit from 1 to 5.
- Apply fixed gates. Featured voices need automatic checks to pass, no applicable blocker, naturalness, pronunciation, and long-form consistency of at least 4, and at least one use-case score of 4. Chinese candidates also need code-switching of at least 3.
July 2026 Evaluation Scope
The voice-quality-phase-b-2026-07-24 run evaluated 18 candidates: nine US English voices, one UK English voice, and all eight available Mandarin Chinese voices. It produced 126 automatic cases and 18 human listening reports. One internal reviewer performed the listening step with headphones in an ordinary room.
The benchmark publishes raw six-dimension scores instead of a cross-language average because English code-switch scores are diagnostic while Chinese code-switch scores are a release gate.
Comparison Page Review
Competitor pages describe product structure rather than pretending to be a permanent price database. Before a review date is published, the competitor's official pricing or product page is checked, VoxFree capabilities are checked against the current application, and exact volatile prices are avoided unless they are necessary and directly sourced. Every page links the official source and states the review date.
VoxFree does not run laboratory head-to-head preference tests against competitor voices. Statements about premium providers being stronger in cloning, emotional delivery, APIs, or enterprise features are product-scope comparisons, not acoustic benchmark results.
Corrections and Updates
When a provider plan, VoxFree limit, model, or quality result changes, the affected page should update its visible review date and the public changelog. Correction requests can be sent to [email protected].