Maya-1

Maya-1

An open-weight text-to-speech model where you describe the voice in plain words and add inline emotion cues. It suits developers and creators who need expressive, designed voices.

Made by maya-research 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: Maya-1 is a speech model with open weights, made by Maya Research, that reads text aloud. Instead of picking from a fixed list of speakers, you write a short description of the voice and how it should sound, then give it the text to speak. It also accepts inline cues for expression. The model makes discrete audio tokens that are decoded into a 24 kHz waveform.

Who it suits: It suits developers and creators who want a designed voice rather than a stock one. The card describes more than 20 emotions and vocal styles.

Title: Who might use it
Industry | Who uses it | How they use it | Example
Games and animation | Character voice designers | Describe a character's age, accent and mood in text | Give a villain a low, gravelly voice with slow pacing
Media | Narrators and podcast producers | Generate narration in a chosen tone | Ask for a calm male narrator with a warm voice
Software | Voice app developers | Replace a fixed speaker menu with described voices | Give an event host voice to an app that reads text aloud

If you only want one plain, fixed voice, you can skip it.

Where to use it: The record lists the transformers library, so Python is the place. The release includes the checkpoint, the tokenizer and reference inference code.

• Python with the transformers library
• Windows, macOS or Linux with Python, to download the files
• Apps that need speech output at 24 kHz

Title: Jobs it does
Use case | Who | What they get back
Character voices | Game and animation teams | Speech in a voice built from a text description
Narration | Producers | Spoken audio in a calm or warm delivery
Expressive lines | App developers | Speech with laughs, cries and other cues inline

How to set it up: You need Python with pip and room for about 6.6 GB of weights in BF16. The reference code also uses a SNAC decoder to turn tokens into audio.

• Python with pip
• The transformers library
• Disk and memory for roughly 6.6 GB of weights

1. Install the Hugging Face download tool.
2. Download the model files to your computer.
3. Follow the reference inference code in the model card.

pip install -U huggingface_hub
hf download maya-research/maya1

How to use it: Write two inputs. The first is a voice description, covering things like age, accent, pitch, pacing and tone. The second is the text to speak, with optional inline cues where you want a laugh or a cry. The model returns speech audio.

Title: The inputs and outputs
Tool | What it does
Voice description | A text prompt that sets the voice and its delivery
Text | The words to speak, with optional inline emotion cues
SNAC decoder | Turns the audio tokens into a 24 kHz waveform

Example: You describe a calm male narrator in his late twenties with a neutral American accent, a warm low voice and a steady pace. You add a short paragraph of text. You get back spoken audio in that voice at 24 kHz.

For newcomers and experts: If you are new, think of it as casting a voice actor with a sentence. Write who the speaker is and how they sound, add one line to read, and listen to the result.

If you know the field, it is a decoder-only model with a Llama architecture and 3.3 billion parameters. It predicts SNAC audio codes that a decoder turns into mono audio. The trade-off against a fixed speaker menu is flexibility in the voice against a larger model to run.

You write a voice description and text
The model predicts SNAC audio tokens
A SNAC decoder rebuilds the waveform
You get 24 kHz mono speech

Good to know: The weights are in BF16. The 4-bit figure below is an estimate from the Hugging Face record, not a file you download.

Title: Approximate weight size
BF16: 6.6 GB
4-bit estimate: 1.7 GB

• License: apache-2.0, with component notes in the card
• Cost to run: none named
• Data it reaches: the model files from Hugging Face, and the text and voice descriptions you give it

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: transformers
• Parameters: 3.3B
• Weights: about 6.6 GB in BF16; about 1.7 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 8,798
• Likes: 943
• License: apache-2.0
• Quantized versions: 12
• Fine-tunes: 15
• Tags: transformers, safetensors, llama, text-generation, voice-design, expressive-speech, audio-generation, text-to-speech

Links

• Model: https://huggingface.co/maya-research/maya1
• Files: https://huggingface.co/maya-research/maya1/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download maya-research/maya1

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