assistant context sentences are getting jumbled up

[a6kme]

pipecat version

0.0.102

Python version

3.13

Operating System

Mac

Issue description

@filipi87 - I believe #3584 has broken something in the way we aggregate context for the assistant. The sentences that the assistant is speaking is getting jumbled up.

I believe it might be because we are setting self._cumulative_time = 0 every time in run_tts method.

pipecat/src/pipecat/services/elevenlabs/tts.py

Line 703 in d99a256

self._cumulative_time = 0

Reproduction steps

Run a pipeline with ElevenLabsTTSService

Expected behavior

The assistant context should not have jumbled aggregation

Actual behavior

The assistant context has jumbled aggregation

Logs

[filipi87]

Hi @a6kme, thank you for reporting this.

Are you able to reproduce the issue using our example ?

• examples/foundational/07d-interruptible-elevenlabs.py

I have tried a couple of times, and it worked as expected every time.

The ElevenLabsTTSService is a TTS service that defines the pause_frame_processing property. This means that the run_tts method should not be invoked until the previous request has finished processing, so self._cumulative_time should not be an issue.

However, if you can provide the steps to reproduce the issue using the example mentioned above, and share the Pipecat logs when it occurs, that would greatly help us understand whether there is still an issue we are missing.

[markbackman]

Here's a fix for the issue:
#3727