Aria TTS

Free plan APIiPhoneSelf-hostedWeb

What Aria TTS is

Aria TTS is a free, open-source text-to-speech model under the MIT license, with about 112 million parameters and English as its listed language. It generates 32 kHz audio and accepts inline tags such as [laughs], [sighs], and [coughs] for expressive cues. Audio streams with word-level timestamps. Deployment options include CPU with ONNX, iOS with ONNX, and GPU, with API, self-hosted, web, and iOS availability listed. On iOS, ONNX inference runs on the device; the site describes this as private because no server round trip is required. Reported CPU performance is 2.7–3.2× realtime on an m6a.large instance. The project says weights are included, with no usage caps or phone-home behavior, and lists commercial use, voice cloning, and API access. It lists $42/mo as infrastructure cost for unlimited use on a reserved m6a.large. Availability remains a caveat because the site says Aria is not quite ready to open the doors yet.

Who it suits

Aria TTS suits developers and teams seeking an open-source English speech model for API, self-hosted, web, CPU, GPU, or iOS deployment. Its MIT license and commercial-use support fit permitted commercial projects.

What it does well

MIT-licensed open-source model
Supports expressive inline speech tags
Streams audio with word-level timestamps
Offers iOS on-device inference
No usage caps or phone-home behavior

What to know first

English is the only listed language
The site says it is not yet ready
Unlimited reserved-instance infrastructure is listed at $42/mo
Model output is 32 kHz audio

Verdict

Aria TTS offers an open model with expressive tags, streaming timestamps, multiple deployment paths, and stated commercial use. Its not-yet-ready status is the main practical constraint despite the free license and included weights.

Facts from Aria TTS's own pages

API access: Yes · aria-tts.org, 20 Sept 2026
Audio output: Aria produces 32 kHz audio using a three-codebook RVQ codec. · aria-tts.org, 13 Oct 2026
Availability: The page says the project is not quite ready to open its doors yet. · aria-tts.org, 13 Oct 2026
Commercial use: Yes · aria-tts.org, 20 Sept 2026
CPU performance: The page reports 2.7–3.2× realtime CPU inference on an m6a.large instance. · aria-tts.org, 13 Oct 2026
Deployment: The listed deployment options are CPU using ONNX, iOS using ONNX, and GPU. · aria-tts.org, 13 Oct 2026
Expressive speech: Inline tags such as [laughs], [sighs], and [coughs] add paralinguistic expressions. · aria-tts.org, 13 Oct 2026
Infrastructure cost: The page lists $42/mo as the total infrastructure cost for unlimited use on a reserved m6a.large. · aria-tts.org, 13 Oct 2026
iOS: Aria runs on iOS using ONNX for on-device inference. · aria-tts.org, 13 Oct 2026
Language: The listed language is English. · aria-tts.org, 13 Oct 2026
License: The project is MIT licensed, and the page says weights are included with no usage caps or phone-home behavior. · aria-tts.org, 13 Oct 2026
Model size: The model has approximately 112 million parameters. · aria-tts.org, 13 Oct 2026
On-device privacy: The site says iOS on-device inference is private and requires no round trip to a server. · aria-tts.org, 12 Oct 2026
Paralinguistic tags: Text can include inline tags such as [laughs], [sighs], and [coughs]. · aria-tts.org, 12 Oct 2026
Platforms: web,ios · aria-tts.org, 20 Sept 2026
Product: Aria is an open-source text-to-speech model licensed under MIT. · aria-tts.org, 12 Oct 2026
Purpose: Aria is an open-source text-to-speech system. · aria-tts.org, 13 Oct 2026
Streaming: It streams audio with word-level timestamps and reports first-chunk latency below 150 ms with a KV-cached reference. · aria-tts.org, 13 Oct 2026
Usage limits: The site says there are no usage caps or phone-home behavior. · aria-tts.org, 12 Oct 2026
Voice cloning: It supports zero-shot voice cloning from reference audio without requiring a transcript. · aria-tts.org, 13 Oct 2026

Where Aria TTS stands on Freedom251

Text-to-Speech Software#29 of 223
Text-to-speech tools#123 of 232