Skip to main content

​
When to Use

  • Best quality, full generation — Kokoro generates all frames at once. Use when you can wait for complete audio before playback.
  • Need streaming/immediate playback — Use PocketTTS instead (~80ms to first audio).

​
Specs

MetricValue
Parameters82M
Voices48
Speed23x RTFx
Peak RAM1.5 GB
ArchitectureFlow matching + Vocos vocoder
PhonemizationeSpeak-NG (GPL-3.0)
Model: FluidInference/kokoro-82m-coreml

​
Quick Start

​
CLI

swift run fluidaudio tts "Welcome to FluidAudio text to speech" \
  --output ~/Desktop/demo.wav \
  --voice af_heart

​
Swift

import FluidAudioTTS

let manager = TtSManager()
try await manager.initialize()

let audioData = try await manager.synthesize(text: "Hello from FluidAudio!")
try audioData.write(to: URL(fileURLWithPath: "/tmp/demo.wav"))

​
Chunk Metadata

let detailed = try await manager.synthesizeDetailed(
    text: "FluidAudio can report chunk splits for you.",
    variantPreference: .fifteenSecond
)

for chunk in detailed.chunks {
    print("Chunk #\(chunk.index) -> variant: \(chunk.variant), tokens: \(chunk.tokenCount)")
    print("  text: \(chunk.text)")
}

​
Pipeline

text → espeak G2P → IPA phonemes → Kokoro model → audio
         ↑                ↑
   custom lexicon    SSML <phoneme>
   overrides here    overrides here
Because espeak runs outside the model as a preprocessing step, you can intercept and edit phonemes before they reach the neural network. This is what enables SSML, custom lexicon, and markdown pronunciation control.

​
Pronunciation Control

Kokoro supports three ways to override pronunciation:
  • SSML tags — <phoneme>, <sub>, <say-as>. See SSML documentation.
  • Custom lexicon — word → IPA mapping files loaded via setCustomLexicon(). See Custom Pronunciation.
  • Markdown syntax — inline [word](/ipa/) overrides in the input text.

​
Kokoro vs PocketTTS

KokoroPocketTTS
Pipelinetext → espeak G2P → IPA → modeltext → SentencePiece → model
Voice conditioningStyle embedding vector125 audio prompt tokens
GenerationAll frames at onceFrame-by-frame autoregressive
Latency to first audioMust wait for full generation~80ms after prefill
SSML supportYes (<phoneme>, <sub>, <say-as>)No
Custom lexiconYes (word → IPA)No
Pronunciation controlFull (phoneme-level)None (model decides internally)
Text preprocessingFull (numbers, dates, currencies)Minimal (whitespace, punctuation)

​
Benchmarks

Same text samples generating 1s to ~300s of output audio, M4 Pro:
FrameworkRTFxPeak RAMNotes
Swift CoreML23.2x1.50 GBLowest memory
MLX23.8x3.37 GB—
PyTorch CPU17.0x4.85 GBKnown memory leak
PyTorch MPS10.0x1.54 GBCrashes on long strings
CoreML matches MLX speed with 55% less peak RAM. PocketTTS benchmarks coming soon.

​
Enable in Your Project

​
Package.swift

dependencies: [
    .package(url: "https://github.com/FluidInference/FluidAudio.git", from: "0.7.7"),
],
targets: [
    .target(
        name: "YourTarget",
        dependencies: [
            .product(name: "FluidAudioWithTTS", package: "FluidAudio")
        ]
    )
]

​
Import

import FluidAudio       // Core (ASR, diarization, VAD)
import FluidAudioTTS    // TTS features