Plattform
ScaiWave ScaiGrid ScaiCore ScaiBot ScaiDrive ScaiKey Modelle Tools & Services
Lösungen
Organisationen Entwickler Internet Service Provider Managed Service Provider AI-in-a-Box
Ressourcen
Support Documentation Blog Downloads
Unternehmen
Über uns Forschung Karriere Investieren Kontakt
Anmelden

Read messages aloud

json
1
2
3
4
5
6
{
  "title": "Read messages aloud",
  "audience": "power_user",
  "summary": "Have any chat message spoken aloud via TTS — with chunked streaming for long messages.",
  "sort_order": 11
}

Read messages aloud#

Any message in a chat can be read aloud by clicking the speaker icon in the message action toolbar. Long messages stream audio in chunks — you hear the first sentence within a second, even for paragraph-length replies.

How to use#

  1. Hover over any message in the chat timeline.
  2. In the action toolbar that appears, click the 🔊 speaker icon.
  3. The icon changes to a spinning loader while audio is fetched, then to a stop square while playing.
  4. Click the stop square to stop playback at any time.

How it works#

ScaiWave splits the message into sentence-sized chunks and synthesizes them in parallel via ScaiSpeak. Playback starts as soon as the first chunk is ready — you don't wait for the whole message. Subsequent chunks queue seamlessly.

For multi-chunk messages, the tooltip shows progress ("Stop reading (2/5)").

Voice selection#

The voice used for read-aloud follows the same resolution chain as voice calls:

  1. Your personal default (Settings → Voice)
  2. Tenant default (set by admin)
  3. First available voice

Caching#

Clicking the same message again replays from cache — no re-synthesis needed within the session. Changing your default voice clears the cache so new playback uses the new voice.

Style consistency#

A consistent style instruction is applied to every chunk:

"Calm, consistent conversational tone. Natural pacing. Maintain the same style across sentences as a continuous passage."

This prevents the TTS engine from varying its delivery between sentences of the same message.

Markdown stripping#

Messages with markdown formatting (bold, italics, headings, bullet points) are stripped to plain text before synthesis. The TTS engine never reads "asterisk asterisk" aloud.

Where to go next#

Updated 2026-05-27 22:31:38 View source (.md) rev 1