Try AnyToSpeech
Benchmark › Speech models › AnyToSpeech Standard voices
AnyToSpeech · our own product · tested 2026-11-22

AnyToSpeech Standard voices

98.9overall · rank 2 of 19 speech models
API model: site pipeline on gemini-3.8-flash-tts Reached through AnyToSpeech pipeline Voice: Kore24 kHz

This is our own product

These are AnyToSpeech's Standard voices, called the way the site calls them: the text goes through our pipeline, which splits long text at sentence ends and joins the pieces, adds number-reading hints, detects the language and retries failed pieces. The voice model underneath is Gemini 3.8 Flash TTS. Measured by us with the same harness and rules as every other row, with nothing tuned for the test.

Profile

Each spoke is one test, 0 at the centre and 100 at the edge.

AnyToSpeech Standard voicesMedian of all models (dashed)

Test by test

The dot shows where this model sits among all 19: left is last, right is first.

Long text

One document sent at five lengths, a single request each. Longest clean read: 15,000 characters.

1,000 characters2,000 characters4,000 characters8,000 characters15,000 characters
10010099.699.999.8

Numbers, dates and symbols

59 of 60 sentences read correctly. Hover a square for the sentence.

Text: Prices rose 3-5% this year.
Heard: Prices rose 3.5% this year.

Emotion control

38 of 40 directed clips measurably changed. Control used: text instruction.

Angry
8/10
Excited
10/10
Scared
10/10
Whisper
10/10

Other languages

Audio from the test

Unedited clips exactly as the model returned them.

Plain sentence
TextThe ferry leaves the harbor every morning just after sunrise, and most of the passengers are still half asleep.
Repetition
TextNo, no, no, no, no. I said no.
Money amount
TextThe total comes to $1,250.50 after tax.
Hard text: percent-range (failed)
miss
TextPrices rose 3-5% this year. HeardPrices rose 3.5% this year.
Spanish
Japanese
Neutral reading
Asked for: angry
Asked for: excited
Asked for: sad (not scored yet)
Asked for: whisper

Compare with

Think a score is wrong? Check it.

Every test item, scoring rule and known limit is published, and the audio behind each number is on the page.

Read the methodology