Translation Update: Optional TTS Termination for Text-Only Pipelines

2026-05-15 · Update

Translation pipelines just got more flexible — text output can now optionally terminate without generating TTS audio, enabling lower-latency text-only workflows.

What Changed

Previously, MediaSFU's translation pipeline always followed the full path: speech → text → translation → TTS audio. The audio synthesis step was always executed, even when your application only needed the translated text.

Now, the TTS (audio) output path is optional. You can configure translation to terminate on the text part, skipping audio synthesis entirely.

Before

Speech → STT → Translation → TTS (always)

• Audio always generated
• Higher latency
• Higher compute cost

After

Speech → STT → Translation → Text ✓ (TTS optional)

• Audio only when needed
• Lower latency for text paths
• Reduced cost

When to Use Text-Only Termination

Live Captions / Subtitles

Display translated text as captions in real-time without generating audio. Perfect for webinars and large meetings where participants read subtitles.

AI Processing Pipelines

Feed translated text into downstream AI systems (summarization, sentiment analysis, entity extraction) without the overhead of audio synthesis.

Meeting Transcription

Generate multilingual transcripts where you only need the text. This is the foundation for the new AI Notes feature.

Low-Latency Requirements

When every millisecond matters, skipping TTS synthesis saves 200-500ms per translation turn.

How to Configure

In your translationConfig, set the output to text-only mode:

// Translation config — text-only output (no TTS)
const translationConfig = {
  sourceLanguage: "en",
  targetLanguage: "es",
  outputMode: "text",  // "text" = no TTS, "audio" = with TTS, "both" = text + audio
  // ... other config options
};

This change is backward-compatible. Existing configurations that don't specify outputMode will continue to generate both text and audio as before.

Explore Translation Options

Configure text-only or full audio translation pipelines for your use case.