Audio clipping on evi-python-quickstart

[NpappaG]

Hi all, hope this is the right place. I got going fast with the python EVI quickstart. However, there was intense clipping of the audio responses no matter what I tried.

Environment:

• macOS Sequoia 15.5 on M1 MBP
• Python 3.11.5 (consistent in pyenv and venv)
• Hume SDK version 0.10.1
• ffmpeg version 7.1.1 (installed via Homebrew)
• sounddevice version 0.4.7

Audio Setup:

• Using MacBook Pro built-in microphone and speakers
• Audio configured through Core Audio

Development Environment:

• Tested with multiple environment managers (miniconda3, uv, & poetry)
• Final test using clean venv as per quickstart guide
• All dependencies installed fresh in isolated environment
• Using fish shell

Steps to reproduce:

1. Created fresh venv: python -m venv evi-env
2. Activated environment: source evi-env/bin/activate.fish
3. Installed dependencies: pip install "hume[microphone]" python-dotenv
4. Ran the example quickstart.py code
5. Successfully connects and transcribes speech
6. Bot responds and audio is received
7. Audio playback consistently clips/cuts out during responses

Issue occurs consistently across all tested environment setups and persists even with fresh installation following the quickstart guide exactly.

Hope that's helpful! LMK if any other details needed.

Best,
Nick Pappageorge

[linear]

HUME-13544 Audio clipping on evi-python-quickstart

[tjasmin111]

Actually, I am facing the same issue of audio disconnecting and clipping in the Quickstart demo:
https://github.com/HumeAI/hume-api-examples/blob/main/evi/evi-python-quickstart/quickstart.py

When I run the demo, the assistant's audio is clipped and I hear it with some pauses in the middle.
I'm on Windows. Can you please fix the issue?

PS C:\Users\Downloads> python .\hume_ai_quickstart.py
WebSocket connection opened.
Configuring socket with microphone settings...
Microphone connected. Say something!
[12:16:12] <chat_metadata> Chat ID: xxxx
...
...

[xssninja]

Me as well. On Mac M3.
For me the audio is clipped or is coming in chunks and isn't buffered for playback. If I wait for the whole audio and reassemble it then it's playable but another problem happens whereby the AI hears itself and responds to itself before it can hear the user and it gets into a feedback loop of only hearing itself.

[twitchard]

The playback issues should be fixed by HumeAI/hume-python-sdk#352. This also includes a fix for allow_user_interrupt=False which wasn't being respected.

The ideal fix for "EVI hears itself" would be adding echo cancellation like we do for the web SDKs. In the meantime the quickstart will prevent the echo loop via allow_user_interrupt=False which is a better experience for users who aren't wearing headphones.