Inconsistencies in the audio RTP node configuration

[GaijinKa]

Description

There are a couple of inconsistencies in the audio RTP node configuration that should be addressed:

1. Channel configuration mismatch

The FFmpeg pipe output channels are explicitly defined, but the _get_waveform method subsequently converts the audio to mono regardless of the channel configuration. This creates unnecessary complexity:

• If the pipeline always processes mono audio, the channels parameter should be hardcoded to 1 and the stereo-to-mono conversion logic in _get_waveform can be removed.

Suggestions

Hardcode channels=1 throughout

2. Hardcoded audio codec

The codec is currently hardcoded to pcm_s16le, which works well for Whisper. However, other models might require different audio formats (e.g., different sample formats or encodings).

Suggestions

Consider making the codec configurable to support future model integrations

---

⚠️ Important: While this task is straightforward, it has the potential to introduce breaking changes. We must:

• Maintain pcm_s16le as the default codec
• Ensure the channels parameter is consistently set to 1 throughout the node codebase