Free trial
No paid tier exists, so a trial is unnecessary: the software is free to download and self-host. An optional Google Colab notebook lets you generate sample speech in Google's environment without a local install.
Learn moreSpeakCrate
Self-hosted Kokoro text-to-speech behind an OpenAI-compatible API
Advanced API
What is SpeakCrate?
SpeakCrate is an AI narration tool. Open-source, MIT-licensed text-to-speech API that runs Kokoro locally in Docker. Offers 54 multilingual voices, six output formats, sentence-level streaming, and an OpenAI-compatible speech endpoint. CPU or NVIDIA GPU, optional offline operation.
SpeakCrate is an open-source text-to-speech server that runs on your own hardware and exposes speech generation through a single REST endpoint. The project wraps Kokoro, a local speech model, and packages it as a Docker image that accepts a POST request to /v1/audio/speech. Text goes in, audio comes back, and nothing leaves the machine you control. There is no OpenAI account involved and no per-character metering. The voice catalogue is the practical draw. SpeakCrate ships with 54 voices across nine languages: American and British English, Japanese, Mandarin Chinese, Spanish, French, Hindi, Italian, and Brazilian Portuguese. Output can be written to MP3, WAV, FLAC, Opus, AAC, or PCM, and requests can stream audio as each sentence is synthesized, either as raw bytes or as Server-Sent Events for clients that prefer event framing. Deployment is deliberately narrow. A container runs on amd64 or arm64 CPUs with roughly 1.5 GB of RAM free, and the model (about 320 MB) downloads on first boot. An optional :cuda image adds NVIDIA acceleration on amd64, but a GPU is never required. A fresh install generates its own API key, which you fetch from inside the container. Persistent volumes keep models across restarts, and with the right files cached plus KOKORO_LOCAL_ONLY enabled the server runs fully offline. The OpenAI compatibility is a migration convenience rather than a full clone. Compatible SDKs and apps can point at SpeakCrate unchanged, and OpenAI voice names map onto Kokoro equivalents, but some options are accepted and ignored (voice instructions among them). That places SpeakCrate alongside other local TTS servers such as openedai-speech, and it pairs naturally with ScribeCrate when a stack needs speech input as well as output. It is aimed at developers, homelab operators, and privacy-conscious product teams, not at people who want a browser tab that reads text aloud.
How do you use SpeakCrate?
1Decide between the Colab demo and a local install
If you only want to hear the output, open the Google Colab notebook and choose Runtime then Run all. For production or private use, continue with a Docker host running Linux with at least 1.5 GB of RAM.
2Start the container
Run the docker run command with a named volume, restart policy, and a port bound to 127.0.0.1. The Kokoro model downloads on first start, so tail the logs and wait for the ready message.3Fetch the generated API key
Fresh installs create and store an API key automatically. Retrieve it with docker exec kokoro kokoro_manage --getkey and keep it in your shell variable for the requests that follow.4Generate your first audio file
Send a POST to /v1/audio/speech with a Bearer token, a JSON body containing model, input, and voice, and an output path. Open the resulting MP3 to check the voice and speed.5Review the full setup guide for production use
The GitHub readme covers GPU images, voice selection, streaming options, authentication, HTTPS, and environment variables for offline and usage-count settings. Work through it before exposing the API beyond the host.
Pros and cons
Pros
- MIT licensed and self-hosted, so text never leaves your own infrastructure.AI
- OpenAI-compatible endpoint lets existing SDKs and apps switch over with minimal code changes.AI
- 54 voices across nine languages, with six output formats including Opus and FLAC.AI
- Streams audio sentence by sentence via raw bytes or Server-Sent Events.AI
- Runs on modest CPU hardware (about 1.5 GB RAM) with GPU acceleration optional, not mandatory.AI
Cons
- No hosted service: you supply the server, the Docker knowledge, and the uptime.AI
- Model download on first start means the initial request waits on roughly 320 MB of network traffic.AI
- OpenAI voice instructions are accepted but ignored, so prompt-driven voice styling is unavailable.AI
- No documentation site beyond the GitHub readme and FAQ, which leaves support to your own reading.AI
- Optional aggregate usage counts are on by default and must be disabled with a separate environment variable for a fully private setup.AI
How much does SpeakCrate cost?
Pricing
The SpeakCrate project is free and open source under the MIT License; you pay only for the hardware or cloud server you run it on. Kokoro is Apache 2.0 licensed, and bundled dependencies keep their own licenses. The optional Self-Hosted AI Builder's Guide ebook is listed at $0.99.
Learn moreSupport
Setup instructions and FAQ live on the homepage, with fuller documentation, an API reference, and environment variable details in the GitHub repository readme. There is no dedicated commercial support channel mentioned.
Learn moreIntegrations
Compatible OpenAI SDKs and apps can call the speech endpoint directly. It also deploys as part of the Self-Hosted AI Stack, and pairs with ScribeCrate for speech-to-text in the same setup.
Features
OpenAI-compatible POST /v1/audio/speech endpoint using Kokoro locally, 54 voices across nine languages, output as MP3, WAV, FLAC, Opus, AAC, or PCM, streaming via audio bytes or Server-Sent Events, CPU (amd64/arm64) or NVIDIA CUDA images, persistent model cache, offline operation with KOKORO_LOCAL_ONLY, and auto-generated API keys.
Learn moreFrequently asked questions about SpeakCrate
How much does SpeakCrate cost?
The SpeakCrate project is free and open source under the MIT License; you pay only for the hardware or cloud server you run it on. Kokoro is Apache 2.0 licensed, and bundled dependencies keep their own licenses. The optional Self-Hosted AI Builder's Guide ebook is listed at $0.99.Does SpeakCrate offer a free trial?
No paid tier exists, so a trial is unnecessary: the software is free to download and self-host. An optional Google Colab notebook lets you generate sample speech in Google's environment without a local install.Does SpeakCrate have an API?
Yes, SpeakCrate offers an API.How do you use SpeakCrate?
The walkthrough on this page covers 5 steps: 1. Decide between the Colab demo and a local install 2. Start the container 3. Fetch the generated API key 4. Generate your first audio file 5. Review the full setup guide for production use.What platforms does SpeakCrate support?
SpeakCrate is available on Web App, Linux, MacOS and Windows.What does SpeakCrate integrate with?
Compatible OpenAI SDKs and apps can call the speech endpoint directly. It also deploys as part of the Self-Hosted AI Stack, and pairs with ScribeCrate for speech-to-text in the same setup.
Status
StatusActive
Views0
Outbound clicks0
Added10/22/2026
Platforms
Web AppLinuxMacOSWindows
Pricing
Free
