NeuroDynamic.TechPay-per-use audio APIs

NeuroDynamic audio & AI · for agents & applications

A voice for your app. AI for your workflow.

Narration, transcription, private voice cloning and text generation for your app or agent. Hear the voices, compare the models and see the price before paying. No subscription required.

Listen to the voicesConnect your agent

USDC on Base · No subscription · Clear price before payment

Hear an example

Emma · British English

Clear, even delivery for an explanation or spoken update.

0:00 / 0:00

Free sample · No wallet needed to listen

0.005 USDC per MP3

Your text, up to 1,000 characters. Seven voices. No account or voice approval.

Compare all services →

Choose the work you need

What would you like to make?

No account or voice approval

Text to speech

0.005 USDC

1,000 characters · MP3

Choose one of seven English voices for spoken updates, instructions or app narration.

Read the narration guide →
Approved account and voice

Short voice cloning

0.01 USDC

1,000 characters

Create a greeting, announcement or application voice with your approved reference.

Read the cloning workflow →
Approved account and voice

Long private narration

0.05 USDC

2,000 words AND 12,000 characters

Turn an article, lesson or narration script into one joined English WAV.

Request cloning access →

Introductory prices per completed job. Long narration must fit both input limits. Short cloning output: 85 seconds / 4 MiB. Long output: 20 minutes / 64 MiB. Check the job’s quote before signing.

Text generation · public beta · age 18+

Uncensored LLMs, paid per request.

Two reduced-refusal models for writing, chat and text tasks. Access through NanoGPT and Abliteration.ai, without opening your own provider account. Lawful-use rules still apply.

Abliterated Model

0.020 USDC per request

A starting point for writing and general text tasks.

Up to 2,000 UTF-8 input bytes · 512 output tokens · Text in JSON

Abliterated Model Large

0.035 USDC per request

A larger model for writing and reasoning tasks.

Up to 2,000 UTF-8 input bytes · 512 output tokens · Text in JSON

No prompt or response retention on our servers. Content is processed temporarily in memory. Third-party policies apply separately; payment and usage metadata are retained. We cannot resend a lost answer, but you can check payment status without another charge.

Introductory fixed prices. An answer can stop at the token limit. No streaming, file uploads or tool execution. Uncensored does not mean every request will be answered or that answers are reliable.

Read the LLM quickstartLive model catalogueLLM termsLLM privacy

For agents and developers

One clear first step.

Using an AI assistant?

Give it this brief. It can check the service and explain what you need before you pay.

Read the brief before copying

Help me assess NeuroDynamic audio and LLM APIs for my task. Read https://api.neurodynamic.tech/llms.txt and https://api.neurodynamic.tech/.well-known/ai-services.json first. Choose catalogue narration for ready-to-use English speech, transcription for a short audio clip, approved private-voice cloning for a custom voice, or an LLM for text generation. If my assistant supports local MCP tools, read https://x402.neurodynamic.tech/integrations/mcp for the free-by-default client; do not install or enable spending without my authority. Check the current limits, price, network and access requirements. You may inspect documentation and an unpaid quote. Do not generate, enrol a voice, sign a payment or spend funds without my authorisation. Keep wallet keys local. If a payment is uncertain, preserve the original request and proof rather than signing again. Explain the next step in plain language.

Your assistant needs web access; paid use also needs an x402-compatible wallet client. Copying this brief does not install anything or authorise spending.

Does your assistant support MCP?

Connect our local client to explore services and prices for free. Optional purchases use your own wallet and a spending cap.

Set up the MCP client →
Reusable integration reference →

Calling the API yourself?

Choose a free quote example

0.005 USDC · up to 1,000 characters · MP3

curl -i -X POST https://api.neurodynamic.tech/v1/audio/speech \
  -H 'Content-Type: application/json' \
  --data '{"input":"Your next update, spoken clearly.","voice":"bf_emma"}'

This example asks for a free price quote. It does not generate audio or make a payment. Expect HTTP 402: a price offer, not a charge.

Full request and payment guideOpenAPI schemaPython narration client

Already integrating transcription or private cloning? The versioned Python starter covers transcription plus short and long private cloning, with explicit payment and saved-proof recovery. Catalogue MP3 narration uses the separate client above.

Python starter ZIPChecksumAgent instructionsService catalogue JSON

Put it to work

A useful addition to your workflow.

A spoken app update

Your application supplies a short update. Choose a catalogue voice and get an MP3 to play.

Text → chosen voice → MP3

A voice note you can search

Send a short recording. Get a transcript your application can save, search or pass to another tool.

Audio clip → transcription → text

An article in your own voice

Enrol an authorised private voice, then reuse it for approved short or long narration jobs.

Approved voice + script → one WAV

These are integration ideas. Your application handles playback, storage and any later processing. This page offers listening samples and API setup; it is not a browser recording editor or checkout.

Start without an account

Find a voice that fits.

Seven English voices. Your text, up to 1,000 characters. Introductory 0.005 USDC per MP3; no reference recording or approval.

Choose an English accent

7 voices shown. Samples are free to play.

Hazel

American English

Informative and instructional.

Suited to: instructions and general narration.

0:00 / 0:00
Voice ID for your requestaf_heart

Bella

American English

Friendly and conversational.

Suited to: assistant updates and general narration.

0:00 / 0:00
Voice ID for your requestaf_bella

Emma

British English

Even and informative.

Suited to: explanations and general narration.

0:00 / 0:00
Voice ID for your requestbf_emma

Alice

British English

Informative newsreader delivery.

Suited to: briefings and general narration.

0:00 / 0:00
Voice ID for your requestbf_alice

George

British English

Soft and informative.

Suited to: guides and general narration.

0:00 / 0:00
Voice ID for your requestbm_george

Michael

American English

Soothing and informative.

Suited to: guides and general narration.

0:00 / 0:00
Voice ID for your requestam_michael

Ben

American English

Friendly assistant delivery.

Suited to: assistant updates and general narration.

0:00 / 0:00
Voice ID for your requestam_fenrir
Use a voice in your app or agentVoice metadata for agents →

Descriptions reflect human listening, not adjustable emotion settings. Text affects pace and expression. Catalogue voices use hexgrad’s Apache-licensed Kokoro model.

Listen before you integrate

The full recording, not just a sound bite.

Four complete 2,000-word demonstrations, reviewed by a human for clarity, pacing and consistency. Each uses the same passage and a synthetic voice reference.

These long demonstrations use synthetic references in the custom cloning service. The ready-to-use catalogue above is a separate MP3 narration service with its own voice IDs and 1,000-character limit. Similarity and accent retention vary.

Emma

2,000 words · 10:55

Clear, even and informative

British female reference

0:00 / 0:00
Compare reference and shorter example

Reference

0:00 / 0:00

Shorter example

0:00 / 0:00

George

2,000 words · 12:04

Soft, measured narration

British male reference

0:00 / 0:00
Compare reference and shorter example

Reference

0:00 / 0:00

Shorter example

0:00 / 0:00

Bella

2,000 words · 11:03

Friendly and conversational

American female reference

0:00 / 0:00
Compare reference and shorter example

Reference

0:00 / 0:00

Shorter example

0:00 / 0:00

Michael

2,000 words · 11:28

Soothing, informative delivery

American male reference

0:00 / 0:00
Compare reference and shorter example

Reference

0:00 / 0:00

Shorter example

0:00 / 0:00

Private voice cloning workflow

Submit. Follow progress. Collect.

  1. 01

    Submit your script

    Choose short or long, and supply your approved private voice ID. Keep the returned job ID.

  2. 02

    Follow its progress

    Poll for status and completed sections. Long work saves progress and shares capacity with other jobs.

  3. 03

    Pay and collect

    When ready, check the quote and pay once. Download one WAV. Use the same saved proof for a repeat download.

Voice enrolment comes first. Generation does not require a fresh human review for every recording. Capacity and start times vary; failed generation is not payable.

Know what to expect.

What is x402, in plain language?

It lets software pay for a request. Your app asks for a price, checks it against your budget, then authorises a USDC payment on Base. A free 402 quote does not move money. You need a compatible wallet client to buy a result; no subscription is required.

Can I use this without coding?

You can listen to the samples here. To create your own result, use an application or wallet-capable assistant that can call these APIs. Copy the agent brief above to explore the setup. There is no browser payment or generation form here yet.

Why are these prices described as introductory?

These are early-access prices while the service develops. Current prices and limits are published, and your client should check the quote before every new payment. A low price is not a promise of unlimited capacity.

Can an agent use cloning immediately?

Cloning requires an operator-issued account and an approved voice reference. Uploading a sample does not grant approval. Once approved, that voice can be reused for automated jobs. Current references expire after 30 days and need renewed enrolment. Catalogue narration and transcription need no separate account.

How long does private cloning take?

A 2,000-word English script produced roughly 11–12 minutes of speech in our reviewed examples. Generation is asynchronous and its speed varies. Status includes section progress and an estimate, not a guaranteed completion time. Short jobs have a 30-minute queue deadline; long jobs a two-hour deadline.

What if a cloning job or payment is interrupted?

Completed long-job sections survive a restart. Keep the original job ID and retry with the same saved payment proof. Repeat delivery does not create a second payment. If a payment result is uncertain, contact support before signing anything new.

What happens to private-cloning text and audio?

Job text and working reference are removed after completion, failure, cancellation or expiry. Completed audio normally remains available for one hour; confirmed payment renews that window. Outputs with uncertain payments are held for recovery. Enrolled references have separate retention. Private operational logs use masked identifiers and exclude customer audio and text.

Retention and recovery differ by service. Read the selected service guide and privacy notice before sending content.

Full cloning terms and workflow · Request access or get support · Directory listing