v0.4.0

NVIDIA Pipecat 0.4.0 (3 March 2026)

Added

• Multilingual support for the voice agent
• AI agent deployment skill for WebRTC, WebSocket, and NAT agent examples
• Jetson Thor edge deployment support
• OpenTelemetry tracing support for ASR, TTS, and LLM services
• Riva Text Filter option to clean and normalize LLM output before TTS processing
• Display of unfiltered transcripts in the WebRTC UI
• Support for Nemotron 3 Nano LLM model

Changed

• BREAKING: Renamed RivaASRService to NemotronASRService and RivaTTSService to NemotronTTSService to better reflect the underlying Nemotron Speech technology. The old names remain available as deprecated aliases.
• Upgraded to pipecat 0.0.98
• Migrated to RTVIObserver for the WebRTC UI
• Changed default TTS sample rate to 22.05 kHz for WebRTC examples
• Updated the Jetson Thor guide to use the public Riva 2.24.0 release

Fixed

• Chat history truncation logic for Nemotron models
• Riva generate_interruptions logic
• TTS chunk cutoff at websocket transport layer by appending silence at the end of each TTS response
• TTS text normalization in NemotronTTSService

Removed

• Riva NMT processor and BlingFire Text Aggregator