Pipecat 1.7 meters every STT audio second

Pipecat 1.7 makes every speech-to-text service report submitted audio seconds and adds PocketTTS, a local CPU-only voice supporting six languages. The framework’s official release notes say continuous services emit usage at each final transcript, segmented services emit per clip, and the measurements flow into client events, logs, and OpenTelemetry spans; the same release adds one local voice path with cloning from a WAV prompt. For builders weighing the lock-in behind proprietary voice-agent pipelines, provider-neutral metering turns speech cost into an observable unit before a multilingual demo becomes a production bill.