phonon-31m-en

phonon-31m-en

A small English text-to-speech model that streams natural 24 kHz audio and runs on a laptop CPU, a phone or in a browser tab. For developers who want speech on the device.

Made by gradium on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: phonon-31m-en is a small English voice model that reads text aloud, one of the Phonon models made by Gradium. It turns text into natural-sounding speech at 24 kHz, in mono.

It starts sending audio while it is still generating the rest. The card calls it light enough for a browser tab, a phone or a laptop processor. It builds on Pocket TTS and the Mimi codec from Kyutai.

Who it suits: Developers who want speech output inside an app without a server, and people who want a light English voice on their own machine.

Industry | Who uses it | How they use it | Example
Software | App developers | Add spoken output on the device | Read a notification aloud
Accessibility | Tool builders | Read text aloud for people who prefer to listen | Speak the text of a page
Content | Creators | Make short English voice clips | Save a line of text as a WAV file

Skip it if you need a language other than English, since this version speaks English only.

Where to use it:

• A command line, through the ptts tool
• Python programs, through the ptts package
• A browser or a phone, through the runtimes the card lists
• Hugging Face Transformers, with a recent source install
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Quick voice clip | Creators | A WAV file spoken in a chosen voice
Streaming speech | App developers | Audio chunks that can play before the full text is done
Local assistant voice | Hobbyists | A spoken reply from a small model on a laptop

How to set it up:

• Python and pip, or the uvx tool that runs the command without installing it
• The ptts package for Python use
• A newer Transformers installed from source, if you want that route

1. Install ptts, or run it once with uvx.
2. Pick the English language option and the 8-bit quantized weights.
3. Choose a voice and give it your text.

uvx ptts --lang en --quant q8 \
  --voice Easton "Hello from Phonon." -o speech.wav

pip install ptts

How to use it: Every runtime takes a language option, which decides how numbers, symbols and dates are read out. This model speaks English, so use en. Pick a voice such as Easton and save the result as a WAV file.

import ptts

tts = ptts.TTS("hf://gradium/phonon-31m-en", lang="en", quant="q8")
tts.save("speech.wav", "Hello from Phonon.", voice="Easton")

Detail | Value
Parameters | 31M, not counting the audio tokenizer
Weights file | 64.5 MB, 8-bit quantized
Frame length | 80 ms
Output | Mono, 24 kHz
Language | English

Title: Size of the weights in MB
8-bit weights file: 64.5 MB
Hugging Face BF16 estimate: 82 MB
Hugging Face 4-bit estimate: 21 MB

Example: You run the command with the Easton voice and the text Hello from Phonon. A moment later you have a file called speech.wav with that line spoken.

For newcomers and experts: If you are new to speech models, run the single command first and listen to the file. Then try the Python version, which can also stream the audio.

If you know the field, a flow-matching transformer predicts the audio latents for each 80 ms step from the text. The decoder of the Mimi codec then turns those latents into sound.

Your text
Flow-matching transformer predicts audio latents
Mimi decoder makes audio
24 kHz speech, streamed or saved

Good to know:

• The Python package can stream, so playback can begin before the whole text has become audio
• Other Phonon models cover French, German, Spanish, Portuguese and all 5 languages together, under a non-commercial license
• All Phonon models share one runtime and file layout, so switching means changing the repository name
• Hugging Face lists 16 quantized and 3 fine-tuned versions made by other people
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download and the text you give it
• Not for: languages other than English

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: transformers
• Parameters: 41.1M
• Weights: about 82 MB in BF16; about 21 MB at 4-bit (estimate)
• Formats: Safetensors, GGUF
• Downloads in the last 30 days: 248,578
• Likes: 160
• License: apache-2.0
• Languages: en
• Quantized versions: 16
• Fine-tunes: 4
• Tags: transformers, safetensors, gguf, gradium_phonon_tts, text-to-audio, text-to-speech, streaming, on-device, phonon

Links

• Model: https://huggingface.co/gradium/phonon-31m-en
• Files: https://huggingface.co/gradium/phonon-31m-en/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download gradium/phonon-31m-en

Then follow the usage in its model card. Nothing in this kit runs by itself.