Orpheus Audio Effects

Orpheus gives you clean, dry speech. OrpheusAudioEffects is the node that makes it sound like it came from somewhere - a phone line, a stadium PA, a cathedral, a chipmunk. It sits between OrpheusGenerate and your preview/save node and applies pitch, speed, gain, reverb, and echo to the generated audio. If you're just narrating text you don't need it; if you're building a character or a scene, it's the difference between "a voice" and "a voice in a place."

The quirk that catches everyone: it doesn't do the processing in Python. It shells out to SoX, a thirty-year-old open-source audio Swiss Army knife, writing your audio to a temp WAV, running SoX on it, and reading the result back. That means unlike the rest of the pack, this node has a real system dependency you have to install yourself.

What it does

The effects run in a fixed order in the code - normalize first, then gain, then pitch, then speed, then reverb, then echo - and each one maps to a SoX effect:

• pitch_shift (-12 to +12 semitones) - shifts pitch via SoX's pitch effect.
• speed_factor (0.5x to 2x) - tempo -s, which changes speed without changing pitch. That's the setting you want for a quickened delivery or a deep, slow drawl.
• gain_db (-20 to +20 dB) with use_limiter (default on) - positive gain gets SoX's limiter flag so you can push volume without clipping.
• normalize_audio (default off) - gain -n, normalizes overall level first. Turn it on if your generated clips vary in loudness between chunks.
• add_reverb with reverb_amount and reverb_room_scale (both 0–100) - room size vs. wetness, same mental model as a synth.
• add_echo with echo_delay (0.1–2 s) and echo_decay (0.1–0.9) - delay is time between repeats, decay is how fast they fade.

Inputs: audio from OrpheusGenerate, the three required sliders above, plus optional sox_path for pointing at your SoX binary. Output: one audio, ready for PreviewAudio or SaveAudio.

The README's preset recipes are worth stealing: phone call (light reverb, no pitch), radio announcer (pitch −2, speed 0.9, gain +3), canyon echo (delay 1.0, decay 0.7).

Installing

The pack itself installs via ComfyUI Manager (search "Orpheus") or git clone https://github.com/ShmuelRonen/ComfyUI-Orpheus-TTS.git into custom_nodes, with pip install ... snac ... for the Python side. SoX is separate:

• Windows: grab the .exe installer from the SoX SourceForge page; default path is C:\Program Files (x86)\sox-14-4-2\sox.exe.
• WSL 2 / Linux: sudo apt-get update && sudo apt-get install sox
• macOS: brew install sox

The node auto-detects SoX in the usual spots - it even checks a list of known Windows paths - and degrades gracefully.

The gotcha to remember

If SoX can't be found, this node doesn't crash. It logs "SoX executable not found", hands your audio back untouched, and you'd swear it worked. That's the trap: you'll dial in a great reverb, hear nothing change, and assume the node is broken. Check the console first. On Windows, if SoX lives somewhere nonstandard, set sox_path to the full sox.exe path. And a production tip: since effects stack in a fixed order, leave gain_db as the last thing you touch - it's the master volume for whatever you already built.