Speech models in Ollama

Ollama can now turn text into speech. The first speech model is Phonon, a family of small, state-of-the-art text-to-speech models from Gradium. Speech models are now available for macOS, with Windows and Linux support coming soon.

ollama run phonon "Hello from Ollama."

Unedited output from Ollama 0.40.11 on a MacBook Pro with M5 Max.

Performance on Apple Silicon

On Apple Silicon, Ollama is accelerated by MLX, Apple’s machine learning framework, for state-of-the-art throughput. On a MacBook Pro with M5 Max, Ollama was 35% faster than the next fastest engine, Gradium’s own runtime.

Throughput

Times faster than real time

Measured on a MacBook Pro with M5 Max, generating 45 seconds of speech. Median of 10 runs. Gradium's own runtime, ptts, uses 8-bit weights on the CPU, fp16 on the Neural Engine and f32 on the GPU; Ollama runs the model in bf16.

Ollama’s implementation is just as accurate, matching the word error rate of the reference implementation in Hugging Face Transformers:

Accuracy

Word error rate

Lower is better. 200 sentences from the Seed-TTS English test set, transcribed with Whisper large-v3.

Voices

Choose a voice with --voice, or voice in the API:

ollama run phonon --voice Hadley "Hello from Ollama."

The voices come from Gradium’s voice catalog. ollama show phonon lists them all.

Languages

Model	Language
`phonon`	English
`phonon:40m-fr`	French
`phonon:40m-de`	German
`phonon:40m-es`	Spanish
`phonon:40m-pt`	Portuguese
`phonon:40m-multilingual`	All five

phonon is released under Apache 2.0. The 40M models are released under CC BY-NC 4.0, which limits them to noncommercial use.

Read replies aloud

Pipe a chat model’s reply into Phonon. Text is spoken as it arrives, so the reply starts playing after its first sentence:

ollama run gemma4 --hidethinking "Tell me a fun fact about owls in two sentences." | ollama run phonon

Run ollama run phonon with no text to type lines at a prompt, or redirect the output to save a WAV file:

ollama run phonon "Save this to a file." > speech.wav

API

Phonon works with the OpenAI-compatible /v1/audio/speech endpoint, so existing code that uses the OpenAI SDK only needs a new base URL.

curl http://localhost:11434/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{
    "model": "phonon",
    "voice": "Zoey",
    "input": "Hello from Ollama."
  }' --output speech.wav

from openai import OpenAI

client = OpenAI(base_url="http://localhost:11434/v1", api_key="ollama")

with client.audio.speech.with_streaming_response.create(
    model="phonon",
    voice="Zoey",
    input="Hello from Ollama.",
    response_format="wav",
) as response:
    response.stream_to_file("speech.wav")

import fs from "node:fs"
import OpenAI from "openai"

const openai = new OpenAI({ baseURL: "http://localhost:11434/v1", apiKey: "ollama" })

const response = await openai.audio.speech.create({
  model: "phonon",
  voice: "Zoey",
  input: "Hello from Ollama.",
  response_format: "wav",
})
fs.writeFileSync("speech.wav", Buffer.from(await response.arrayBuffer()))

The response streams while the audio is generated, so it can play right away:

curl -sN http://localhost:11434/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"model": "phonon", "input": "Playback starts right away."}' \
  | ffplay -nodisp -autoexit -loglevel error -

Audio is 24 kHz mono, returned as WAV by default or as raw 16-bit samples with "response_format": "pcm".

Get started

Speech models run on Apple Silicon Macs. Download Ollama 0.40.11 or later, then pull Phonon:

ollama pull phonon

For more information, see: