uTPro.Feature.TextToSpeech 18.0.1

uTPro.Feature.TextToSpeech

Documentation: https://docs.utpro.dev/uTPro.Feature.TextToSpeech/

Add a "listen to this article" player to any Umbraco 16, 17 or 18 page with one line in a template. You pass the text to read, visitors get an accessible audio player (play/pause, speed, progress, resume, download), and you choose the voice from a backoffice workspace. It is free and offline by default: the Browser voice needs no API key, no files and no outbound traffic. When you want consistent, higher-quality audio, map a culture to a server voice (Piper or Kokoro, fully local, or any OpenAI-compatible speech API). Server audio is generated in the background, cached on disk by content hash and reused, so a page render never waits for a synthesizer.

Part of the uTPro feature family (alongside Simple Cart, Simple Form Builder, SEO Audit and others).

Which package do I need?

Start with the core package. Add an offline engine only when you want a server voice.

The core works on its own: the Browser voice and the OpenAI-compatible provider are both built in. Piper and Kokoro are optional add-ons for local server voices. You install an add-on instead of, not before, the core: each add-on depends on the core package, so dotnet add package uTPro.Feature.TextToSpeech.Piper pulls in uTPro.Feature.TextToSpeech automatically at the same version. If a mapped engine is missing or misconfigured, pages fall back to the Browser voice, so a culture is never silent.

Install

dotnet add package uTPro.Feature.TextToSpeech

That's it. The package self-wires its services, migration and backoffice UI through an Umbraco composer, so there are no changes to your Program.cs. Requires Umbraco 16, 17 or 18 (net9.0 / net10.0).

Optional offline voices are separate packages, so the core stays small:

dotnet add package uTPro.Feature.TextToSpeech.Piper    # fast, local, includes Vietnamese
dotnet add package uTPro.Feature.TextToSpeech.Kokoro   # natural English and a few other languages, local

Use it

In any template, pass the text you want read:

@await Component.InvokeAsync("TextToSpeechPlayer", new { content = Model, text = bodyHtml })

content is the page (for its culture and key), text is what to read (HTML is allowed and stripped). Leave text out and the player falls back to the first non-empty property in Rendering.TextPropertyAliases (bodyText, content by default).

The template decides what is read because article bodies are usually Block List / Block Grid, and only your template knows which blocks are prose.

Voices

Pick the voice per culture in Settings → Advanced Settings → Text To Speech → Voices. If a mapped engine is missing or not configured, the page falls back to the Browser voice, so a language is never silent.

How server audio works

1. The template renders the player. The package normalises the text, applies your pronunciation rules and computes a content hash.
2. It registers that hash in a database table and queues generation in the background. The page does not wait.
3. The player streams /api/text-to-speech/audio/{hash}. The file is generated once, stored under umbraco/Data/uTProTextToSpeechCache, and served with ETag and range support.
4. Publishing a page marks its audio stale; the next render produces new audio and the old file is purged after OrphanRetentionDays (7).

Backoffice

Under Settings → Advanced Settings → Text To Speech:

• Voices - engine status cards, a voice per culture, preview.
• Pronunciation - replacement rules ("TP.HCM" becomes "Thành phố Hồ Chí Minh") with a before/after test box.
• Audio cache - every generated file with search, filters, play, download, regenerate and delete.

Every document also gets an Audio content app showing the audio of that page.

Configuration

Everything has a working default. Settings live under uTPro:Feature:TextToSpeech:

{
  "uTPro": {
    "Feature": {
      "TextToSpeech": {
        "Provider": "Auto",
        "OpenAi": { "Endpoint": "https://api.openai.com", "ApiKey": "", "Model": "gpt-4o-mini-tts" },
        "Piper": { "ExecutablePath": "", "ModelsRoot": "" },
        "Kokoro": { "ModelPath": "", "WarmUp": true }
      }
    }
  }
}

Keep ApiKey in user-secrets or environment variables, not in a committed file. The full reference is in the documentation.

Security

• The public audio endpoint accepts only a hash your own template registered. Visitors cannot send text, so the site cannot be used as a free text-to-speech service.
• Calls to the speech API are SSRF-guarded at connect time: loopback, private, link-local and reserved addresses are refused unless you opt in with OpenAi:AllowLocalEndpoint.
• The API key is never returned by any endpoint; the backoffice shows only that one is set.
• Piper is started with an argument list (never a shell), the voice id must be a bare file name inside the models folder, and every run has a timeout.
• Backoffice endpoints are authenticated; the Voices, Pronunciation and Audio cache tabs need Settings access, the Audio content app needs Content access.

Building / packing

dotnet build uTPro.Feature.TextToSpeech.sln -c Release
dotnet test tests/uTPro.Feature.TextToSpeech.Tests/uTPro.Feature.TextToSpeech.Tests.csproj -c Release
pwsh ./pack.ps1

pack.ps1 wipes bin/obj, packs the core and both add-ons into Build/, evicts the version from the local NuGet cache and mirrors the docs to the public repo.

Test site

src/uTPro.Feature.TextToSpeech.TestSite is a minimal Umbraco 18 host (SQLite, unattended install as admin@example.com). Run it yourself with dotnet run --project src/uTPro.Feature.TextToSpeech.TestSite, create a document type with a bodyText rich text property and a template that calls the player, then publish a page.

License

See LICENSE.txt. Free to use in your own Umbraco sites, including commercial ones.