Every English Kokoro Voice, With Audio Samples

Kokoro is a small open-source text-to-speech model (82 million parameters, Apache 2.0) with 54 voices across nine languages. Most voice lists online give you the names and nothing else. This page plays all 28 English voices, American and British, reading the same passage, with their official quality grades and a few measurements, so you can choose by listening instead of guessing from a name.

Contents

How to read this page
Which voices to use
All 28 English voices
What the measurements show
Using the voices

How to Read This Page

Every sample reads the same 47-word passage, a short story beat with a number, a question and a change of mood:

On the twelfth morning he found footprints in the wet sand below the tower. They came up from the waterline, circled the building twice, and stopped at the door. He told himself it was a seal. Then he noticed they were barefoot. Who walks out of the sea at three in the morning?

For each voice the table shows:

Grade: the overall grade from Kokoro's author in the official voice list. In their words, the grades "are intended to be estimates of the quality and quantity of its associated training data, both of which impact overall inference quality." A is best and F is worst.
Words per minute: how fast the voice read this passage, pauses included.
Median pitch: a rough measure of how high or low the voice sits. As a guide, most adult male voices fall around 85 to 155 Hz and female voices around 165 to 255 Hz.

All samples were generated locally with the kokoro Python package (0.9.4) and set to the same loudness, so differences in volume are not differences in the voice. Kokoro produces 24 kHz audio; our audio formats guide shows what that means for the file you export.

Which Voices to Use

Start with these:

af_heart (grade A) and af_bella (A-): the two highest-graded voices. A steady pace with a pause at the end of each sentence. The safe default for narration.
bf_emma (B-): the best British voice by a distance.
af_nicole (B-): much slower than every other voice, with long pauses. Worth trying for calm, close-up narration, too slow for most short-form video.

For a male narrator, Kokoro is weaker. No American male voice grades above C+; the best are am_michael, am_fenrir and am_puck. The best British male voices, bm_george and bm_fable, grade C. If a male voice is central to your channel, compare these against a preset or designed voice from a larger model first.

Treat anything graded D or below as a last resort. am_adam is graded F+, the lowest of all 28.

All 28 English Voices

Ordered by grade, best first.

Voice	Accent	Gender	Grade	Words per minute	Median pitch
`af_heart`	American	Female	A	198	201 Hz
`af_bella`	American	Female	A-	189	207 Hz
`af_nicole`	American	Female	B-	125	161 Hz
`bf_emma`	British	Female	B-	214	185 Hz
`af_aoede`	American	Female	C+	200	181 Hz
`af_kore`	American	Female	C+	201	148 Hz
`af_sarah`	American	Female	C+	201	199 Hz
`am_fenrir`	American	Male	C+	204	132 Hz
`am_michael`	American	Male	C+	185	118 Hz
`am_puck`	American	Male	C+	210	106 Hz
`af_alloy`	American	Female	C	233	146 Hz
`af_nova`	American	Female	C	234	157 Hz
`bf_isabella`	British	Female	C	215	203 Hz
`bm_fable`	British	Male	C	233	116 Hz
`bm_george`	British	Male	C	194	140 Hz
`af_sky`	American	Female	C-	215	152 Hz
`bm_lewis`	British	Male	D+	190	82 Hz
`af_jessica`	American	Female	D	237	213 Hz
`af_river`	American	Female	D	243	170 Hz
`am_echo`	American	Male	D	229	108 Hz
`am_eric`	American	Male	D	247	160 Hz
`am_liam`	American	Male	D	235	131 Hz
`am_onyx`	American	Male	D	232	86 Hz
`bf_alice`	British	Female	D	216	214 Hz
`bf_lily`	British	Female	D	216	207 Hz
`bm_daniel`	British	Male	D	229	122 Hz
`am_santa`	American	Male	D-	203	141 Hz
`am_adam`	American	Male	F+	217	118 Hz

What the Measurements Show

Higher-graded voices pause; lower-graded ones rush. Apart from the deliberately slow af_nicole, the A and B voices read the passage at about 190 to 215 words per minute with three pauses at sentence ends. The voices graded D read it at about 215 to 250 words per minute with one pause or none, so a dramatic beat like the final question has no room to land. If a lower-graded voice is the one you like, write shorter sentences and add explicit pauses.

Some voices may stumble on numbers. When we transcribed every sample with Whisper, "twelfth" and "three" did not come through clearly from af_aoede and bm_daniel. Listen to their samples before using them for scripts with a lot of numbers.

Pitch varies more than the names suggest. The female voices range from 146 Hz (af_alloy) to 214 Hz (bf_alice), the male voices from 82 Hz (bm_lewis) to 160 Hz (am_eric).

Using the Voices

In InkSlop, pick a voice in the creator, or set it in a script:

[tts engine="kokoro_tts" voice="bf_emma"]
  On the twelfth morning he found footprints in the wet sand below the tower.
[/tts]

British voices are read with British pronunciation automatically, because the language is taken from the voice ID. For other languages, set lang on the narration (for example lang="es") and a native voice is chosen. Kokoro is one of the cheaper voices in InkSlop at 0.5 credits per 1,000 characters; see what 100 videos cost.

In your own code, the first letter of a voice ID is its language (a American, b British, e Spanish, and so on), and the pipeline must use the same code. Pairing a British voice with the American pipeline gives it American pronunciation:

from kokoro import KPipeline
import numpy as np, soundfile as sf

voice = "bf_emma"
pipeline = KPipeline(lang_code=voice[0])   # "b" for British voices
audio = np.concatenate([r.audio for r in pipeline("Hello from the lighthouse.", voice=voice)])
sf.write("hello.wav", audio, 24000)

To compare Kokoro with Qwen3, MiMo and ElevenLabs on the same passage, see the voice comparison. For running Kokoro on a server, including how fast it is on CPU, see Kokoro in production.