Neuphonic Review, Pricing & Alternatives (2026)

Neuphonic is a London startup founded by Sohaib Ahmad, a former hedge fund quantitative trader who is now chief executive, and Jiameng Gao, previously a co-founder of the AI dubbing company Papercup. The two met while studying machine learning at Cambridge University. In October 2024 Neuphonic raised a £3 million pre-seed round led by Moonfire VC, with Tiny VC, Salica Oryx Fund and Cur8 Capital participating, according to Tech.eu. At that point its API was in closed beta and the company said its streaming method could start speaking in 25 milliseconds. Since then it has opened the cloud API to self-serve customers, published a family of open-weight models on Hugging Face and GitHub, and signed a distribution deal that puts its voices inside the Vapi voice agent platform.

There are three ways to use Neuphonic. The cloud API at app.neuphonic.com streams text-to-speech in more than 40 languages, with a browser playground for testing voices and instant cloning from a few seconds of audio. Developers who need data to stay local can download the NeuTTS models and run them on a laptop, phone, Raspberry Pi or private server with no internet connection. Enterprise customers can have Neuphonic deploy and fine-tune models inside their own infrastructure. A typical voice agent build sends a language model's output to the API word by word, and Neuphonic begins speaking before the full sentence has arrived. Voice agent frameworks including Vapi and LiveKit list Neuphonic as a supported voice provider.

NeuTTS Air, released in 2025, was the company's first open model. It is built on Alibaba's Qwen 0.5B language model plus NeuCodec, Neuphonic's own single-codebook audio codec, and it clones a voice from three to 15 seconds of reference audio. It ships in quantised GGUF formats so it generates speech in real time on mid-range CPUs. Smaller NeuTTS Nano models followed for English, Spanish, French and German. The newer NeuTTS-2M adds cloning in seven languages including Japanese, Korean and Chinese, while NeuTTS-2E offers English speech with seven emotion modes and four fixed speakers. NeuDecide turns spoken commands into tool calls. Every NeuTTS output carries an inaudible Perth watermark so generated audio can be identified later.

Neuphonic's pricing page lists four plans. Free gives personal-use access to every voice and language with 25,000 characters a month, about 30 minutes of speech. Starter costs $5 a month for 250,000 characters (about five hours), one concurrent request, instant voice cloning and a commercial licence. Business costs $499 a month for 30 million characters (about 600 hours), 15 concurrent requests, team management and dedicated support on Discord and Slack. Enterprise is quoted and covers fine-tuned voices, on-device and on-premises deployment, custom concurrency, SLAs and round-the-clock support. The open models carry no usage fees, though licences differ by model: NeuCodec and NeuDecide use Apache 2.0, and NeuTTS-2E uses Neuphonic's own NeuTTS Open License, so check the terms before shipping a product.

Neuphonic is a small company, and its voice library and studio tools are much thinner than ElevenLabs, which also offers dubbing, sound effects and a large community voice marketplace. The gap between the $5 Starter and $499 Business plans leaves growing teams with a large jump. Open-model quality is good for the size but behind the largest cloud models on expressiveness. Cartesia competes most directly, with low-latency cloud voices and its own on-device option, backed by far more funding. Kokoro and Piper are free open-source alternatives for local speech without cloning. Pick Neuphonic when you need voice cloning that runs offline on ordinary hardware, such as in toys, kiosks, healthcare devices or privacy-sensitive apps, or a cheap commercial cloud API for a hobby or early-stage project.