Catalog census · 2026-10
What 1,292 AI voices actually look like
A census of the whole VocalVia catalog: 988 English voices, 189 Chinese, 101 Indonesian; 805 male, 380 female, 93 gender-neutral; 85.4% cloned from a speaker sample, 14.6% original recordings. The dataset is below, free to download and quote.
The dataset in one paragraph
A concentrated catalog with one unusual pocket of neutral voices.
If you only read one paragraph, read this one — it is written so it can be quoted as-is.
As of 2026-10, the VocalVia catalog holds 1,292 voices in 8 languages. English dominates with 988 voices (76.5%), then Chinese with 189 (14.6%) and Indonesian with 101 (7.8%); the remaining five languages total 14 voices. 805 voices are male (62.3%), 380 female (29.4%), 93 neutral (7.2%), and 14 carry no label. Neutral voices sit almost entirely in English — 87 of 93 — so in practice a gender-neutral narrator is an English-only option. 85.4% of the catalog is cloned from a speaker sample; 14.6% are original recordings.
Key numbers
The composition at a glance.
Every figure below is computed from the same CSV, so a reader can reproduce the table exactly.
- Voices in catalog
- 1,292
- Languages
- 8
- Curated / community added
- 1154 / 138
- Gender labels
- 805 male · 380 female · 93 neutral · 14 unlabeled
- Age labels
- 534 middle aged · 411 young · 221 old · 126 unknown
- Cloned / original recordings
- 1103 (85.4%) / 189 (14.6%)
- Distinct tags
- 128
- Median preview clip
- 14s
- Likes across catalog
- 64,019
Languages
Three languages carry 98.9% of the catalog.
A catalog this concentrated changes what “multilingual” means in practice: outside the top three, you are choosing from a handful of voices rather than a range.
| Language | Voices | Share | Male / female / neutral |
|---|---|---|---|
| English | 988 | 76.5% | 617 / 281 / 87 |
| Chinese | 189 | 14.6% | 119 / 60 / 0 |
| Indonesian | 101 | 7.8% | 63 / 35 / 3 |
| Japanese | 8 | 0.6% | — |
| Spanish | 2 | 0.2% | — |
| German | 2 | 0.2% | — |
| Russian | 1 | 0.1% | — |
| Dutch | 1 | 0.1% | — |
The three largest languages are the only ones with enough voices to split by gender. Japanese, Spanish, German, Russian, and Dutch are counted in the CSV but are too small to describe beyond their totals.
Gender
A male-skewed catalog with a small, English-only neutral tier.
Gender labels describe how listeners perceive a voice, not the identity of the person who recorded it. That distinction matters when the label is used as a casting filter.
Neutral voices are concentrated by language: English 87, Japanese 3, Indonesian 3, Chinese 0. A listener who needs a narrator that does not read as male or female has 87 English options and 6 elsewhere.
Age
Most voices are labelled middle-aged.
Age labels describe perceived timbre rather than the speaker's actual age, which is why they are useful for casting and unreliable for anything else.
| Age label | Voices | Share |
|---|---|---|
| Middle Aged | 534 | 41.3% |
| Young | 411 | 31.8% |
| Old | 221 | 17.1% |
| Unknown | 126 | 9.8% |
Origin
85% of the catalog is cloned from a sample.
This is the number most catalogs do not publish, and the one that decides how a voice can be used.
85.4%
Cloned (1103 voices)
Built from a speaker sample uploaded by a user. The voice exists because someone supplied audio; the consent behind that upload is what governs its use.
14.6%
Original (189 voices)
Recorded as a single finished asset rather than derived from an upload. If a project needs a voice with a simple provenance chain, filter the CSV on this column.
Tags
What people label voices with, and why the vocabulary is messy.
Tags are free-form and written by whoever added the voice, so they mix delivery, role, tone, and even age. We publish them as-is rather than cleaning them up.
128 distinct tags cover 1,292 voices. “middle aged” appears both as an age field and as a tag on 454 voices, which is the clearest example of why tag counts should be read as descriptive noise, not taxonomy.
Samples
Preview clips are short, and likes measure browsing.
Two counters that are often read as quality signals, shown here for what they actually are.
14s
Median preview clip
Mean 14.2s, range 5–31s. Long enough to judge timbre and pacing; not long enough to test how a voice holds up over a 20-minute episode.
64,019
Likes across the catalog
Median 9 per voice, one outlier at 5,446. A high count means a voice gets browsed and previewed often, which is not the same as being right for a given script.
Method
How this was counted, and what it does not cover.
- Source. The public catalog as served by the site on 2026-10-29, all 1,292 entries, with no filter on the curated flag.
- Labels. Language, gender, age, and origin are the values published with each voice. 14 voices have no gender label and 126 no age label; they stay visible as “unknown” rather than being imputed.
- Tags. Lowercased and de-duplicated for counting; the original strings are in the CSV so the cleaning can be checked.
- Counters. Likes and clip duration are platform-side values at snapshot time. They move, and this page does not re-measure them.
- Scope. One catalog, one date. Other providers publish different mixes, and most publish nothing — so this is a baseline, not a market survey.
Cite this
A citation line you can paste.
Numbers on a page are easy to disagree with and hard to quote. This block exists so the dataset can be referenced without re-describing it.
VocalVia. (2026, October). AI voice catalog census: 1,292 voices across 8 languages. https://www.vocalvia.com/voice-catalog-report
The CSV behind every figure: vocalvia-voice-catalog-2026-10.csv (1,292 rows, one per voice). Licensed for reuse with attribution.
Frequently asked questions
How many AI voices does the VocalVia catalog contain?
1,292 voices as of 2026-10, across 8 languages. 1154 are curated by us and 138 were added from community uploads. All of them are available in the product, and every one is listed in the downloadable CSV.
How many gender-neutral AI voices are available?
93 voices are labelled neutral, which is 7.2% of the catalog. 87 of them are English. Chinese, which has 189 voices, currently has none labelled neutral.
Are the voices original recordings or clones?
1103 voices (85.4%) are cloned from a speaker sample supplied by a user, and 189 (14.6%) are original recordings. The CSV marks each voice, so you can restrict a project to original recordings if that matters for your use case.
Can I reuse this dataset?
Yes, with attribution. The CSV is a snapshot of the public catalog and is free to cite, quote, or analyse. Cite it as "VocalVia, AI voice catalog census, October 2026" with a link to this page, so a reader can check the numbers against the source.
Does this dataset describe the whole TTS market?
No. It describes one catalog on one date. Language and gender distributions at other providers differ, and most do not publish their composition at all. Treat these numbers as a baseline for comparison, not as a market estimate.
Related guides
Casting a project? Start with the catalog itself.
Every voice in this census is available in the product, and 50 trial credits cover a first episode.