4 filings, one each in 4 layers.
Batch multilingual STT with 8k-token context and 2-for-1 translate lands a week before v4 Real-Time — the async foundation later v5 rows replace.
Soniox released stt-async-v4 for recorded audio, a single universal model covering 60+ languages with native code-switching, deep context up to 8,000 tokens, and simultaneous transcription plus translation in one API call.
Soniox v4 Async (stt-async-v4)
Local TTS paths matter for air-gapped and cost-sensitive cascaded bots.
pipecat-ai 0.0.101 adds PiperTTSService for offline Piper voices and KokoroTTSService for Kokoro-82M, plus Google TTS regional location.
pipecat-ai 0.0.101
Incomplete-turn filtering and pipeline VAD set up the later FilterIncompleteUserTurnStrategies stack in 1.2.0.
pipecat-ai 0.0.101 adds UserTurnCompletionLLMServiceMixin to filter incomplete user turns, plus VADProcessor and VADController for in-pipeline speech detection.
pipecat-ai 0.0.101
Contact-centre Genesys AudioHook becomes a first-party telephony path for Pipecat bots.
pipecat-ai 0.0.101 adds GenesysFrameSerializer for the Genesys AudioHook WebSocket protocol with bidirectional audio.
pipecat-ai 0.0.101