Neuphonic
London text-to-speech startup with an open-weight on-device model that needs no GPU
Quick Overview
| Company | Neuphonic Limited |
|---|---|
| Category | AI Voice & Speech |
| Headquarters | London, United Kingdom |
| Founded | 2024 |
| EU Presence | United Kingdom (adequacy decision, outside the EEA) |
| Data Location | Cloud, on-device or on-premises deployment |
| Open Source | No |
| Pricing | Not published; API playground available to test, contact for API and deployment pricing |
| Replaces | ElevenLabs, Amazon Polly, Google Cloud TTS |
Detailed Review
Neuphonic is the newest and smallest company in this category, incorporated in London in April 2024, and the one built specifically to avoid a GPU.
Its flagship NeuTTS model runs at roughly double real-time speed on two CPU threads, with ultra-fast voice cloning built in, aimed squarely at developers who need text-to-speech inside a product without the cost or latency of a round trip to a cloud GPU. It is a narrower, more technical proposition than the consumer-facing tools elsewhere in this category.
The model supports English, Spanish, German, French, Japanese, Korean and Chinese, and Neuphonic publishes NeuTTS as an open-weight model on GitHub and Hugging Face rather than keeping the architecture entirely closed — a genuine point of difference from every other tool in this category, all of which are closed commercial systems. Deployment options run from a hosted cloud API to on-device execution to secure on-premises installation, the same flexibility Speechmatics offers, built by a company a fraction of its size.
Neuphonic Limited has no published pricing anywhere on its site: evaluation happens through an API playground rather than a quoted plan, and there is no certification list to weigh against Speechmatics' or Synthflow's. For a hobbyist or a small team willing to self-host the open model, that is close to free; for a business wanting a contract and a support line, it currently means an email to sales rather than a checkout page.
What Neuphonic does well
- Open-weight NeuTTS model published on GitHub and Hugging Face
- Runs on 2 CPU threads without a GPU
- Ultra-fast voice cloning built in
- Cloud, on-device or on-premises deployment
- Genuinely open architecture, unusual in this category
Where Neuphonic falls short
- No published pricing anywhere on the site
- Very young company — incorporated in April 2024
- No stated certifications to check against enterprise rivals
- Seven languages, fewer than Speechmatics' 55+ or ReadSpeaker's 50+
Standout feature. The only open-weight model in this category, published on GitHub, running twice as fast as playback without a GPU.
Pros and Cons
Pros
- Open-weight NeuTTS model published on GitHub and Hugging Face
- Runs on 2 CPU threads without a GPU
- Ultra-fast voice cloning built in
- Cloud, on-device or on-premises deployment
- Genuinely open architecture, unusual in this category
Cons
- No published pricing anywhere on the site
- Very young company — incorporated in April 2024
- No stated certifications to check against enterprise rivals
- Seven languages, fewer than Speechmatics' 55+ or ReadSpeaker's 50+
Alternatives to Neuphonic
Frequently Asked Questions
What is Neuphonic?
Neuphonic is the newest and smallest company in this category, incorporated in London in April 2024, and the one built specifically to avoid a GPU.
Its flagship NeuTTS model runs at roughly double real-time speed on two CPU threads, with ultra-fast voice cloning built in, aimed squarely at developers who need text-to-speech inside a product without the cost or latency of a round trip to a cloud GPU. It is a narrower, more technical proposition than the consumer-facing tools elsewhere in this category.
Where is Neuphonic based?
Neuphonic operates from London, United Kingdom, which places it under United Kingdom (adequacy decision, outside the EEA).
What does Neuphonic cost?
Not published; API playground available to test, contact for API and deployment pricing.
Who is Neuphonic best for?
Developers wanting an open-weight, on-device text-to-speech model. The only open-weight model in this category, published on GitHub, running twice as fast as playback without a GPU.
What are the drawbacks of Neuphonic?
No published pricing anywhere on the site. Very young company — incorporated in April 2024. No stated certifications to check against enterprise rivals. Seven languages, fewer than Speechmatics' 55+ or ReadSpeaker's 50+.