Skip to main content
Last Updated: July 13, 2026
Deploy from the SaladCloud Portal.

​
Overview

This recipe runs hexgrad/Kokoro-82M behind an OpenAI-compatible text-to-speech API. Send text to the deployment and receive generated audio in formats including MP3, WAV, FLAC, Opus, AAC, and PCM. Kokoro is an 82 million parameter open-weight TTS model. Its small size makes it useful for cost-conscious speech generation such as product narration, voice agents, accessibility features, audio previews, and batch audio workflows. The server supports more than 50 Kokoro voices and produces 24 kHz audio.

​
Quick Start

  1. Open the SaladCloud Portal.
  2. Deploy the Kokoro TTS API recipe.
  3. Enter a Container Group Name.
  4. Decide whether to enable Require Container Gateway Authentication:
    • Enabled: requests must include your SaladCloud API key.
    • Disabled: anyone with the URL can call the TTS service.
  5. Deploy and wait until the replica is ready.
Startup downloads and initializes the model. It can take several minutes before the health check passes.

​
Defaults

The recipe comes preconfigured with these defaults:
  • Model ID: hexgrad/Kokoro-82M
  • Container image: hwdsl2/kokoro-server:latest
  • Command equivalent: python3 -m uvicorn api_server:app --host :: --port 8880
  • Container port: 8880
  • Default voice: af_heart
  • Default speed: 1.0
  • Default sample rate: 24 kHz
  • Model cache: /var/lib/kokoro
  • Readiness and liveness probe: GET /health
  • Authentication: enabled by default through the Salad Container Gateway
The recipe overrides the image’s startup command so Uvicorn binds to IPv6 host ::, which allows Salad ingress to reach the API. The container’s own bearer-token authentication is disabled with an empty KOKORO_API_KEY; use the Salad Container Gateway authentication setting to control public access.

​
API Endpoints

Useful endpoints include:
  • GET /health - readiness probe and health check
  • GET /v1/models - model information
  • GET /v1/voices - available voices
  • GET /docs - Swagger documentation
  • POST /v1/audio/speech - generate speech audio from text

​
Authentication

Require Container Gateway Authentication is available in the deployment form and is enabled by default.
  • Enabled: every request must include the Salad-Api-Key header.
  • Disabled: anyone with the deployment URL can call the API.
If you enable authentication, see Sending Requests for the header format.

​
Example Request

This request generates a WAV file named kokoro.wav:
If you disabled authentication during deployment, omit the Salad-Api-Key header. The endpoint returns binary audio rather than JSON. For this request, the response has Content-Type: audio/wav. If you omit response_format, the server returns MP3 audio by default. The API accepts kokoro, tts-1, or tts-1-hd in the model field. All three values use hexgrad/Kokoro-82M on this server.

​
Test A Deployment

Check health:
List available voices:
Generate a short WAV file, then inspect its format:
The file command should identify kokoro-test.wav as WAV audio. Add the Salad-Api-Key header to each request if gateway authentication is enabled.

​
Configuration Notes

  • Supported response formats are mp3, wav, flac, opus, aac, and pcm.
  • Language codes include a for American English, b for British English, e for Spanish, f for French, h for Hindi, i for Italian, j for Japanese, p for Brazilian Portuguese, and z for Mandarin Chinese. Leave the language code blank to infer the pipeline from the selected voice.

​
Source Code